Home

Oral History Panel on the Development and Promotion of the Intel

1. Oral History Panel on the Development and Promotion of the Intel 4004 Microprocessor Interviewees Federico Faggin Hal Feeney Ted Hoff Stan Mazor Masatoshi Shima Interviewed by Dave House Edited by David Laws Recorded April 25 2007 Mountain View California CHM Reference number X4024 2007 2007 Computer History Museum Oral History Panel of the Intel 4004 Microprocessor Dave House Welcome to the video history of the MCS 4 also known as 4004 microprocessor development joined Intel in February of 1974 so was not at Intel at the time of the 4004 but later worked with each of these individuals in one way or another So what I d like to do is start out just go through the table here and have everyone give their background up to the point where they started working on this product So Stan could you start Stanley Mazor Thank you Dave My name is Stanley Mazor was raised in Oakland California and attended Oakland public high schools went to San Francisco State in 1960 and was introduced to a computer for the first time during the years 1960 1962 took the two computer courses they had in programming There was no computer science at San Francisco State in those days And one of the projects that had in school was to write an interpreter for a virtual machine called IPL V The IBM 1620 was a decimal machine and was to affect my career and some of the contributions made in subsequent years was a l
2. Hey it may be more pervasive than the fractional horsepower motor and he wasn t even close It was more pervasive than that But again it s making the right decisions and going the right path Ted you were talking about the Tl version of the 8008 the initial spec and the reason they TI were at three chips their initial spec CHM Ref X4024 2007 2007 Computer History Museum Page 34 of 36 Oral History Panel of the Intel 4004 Microprocessor called them to be doing a serial machine And as soon as they found out that Intel was doing a parallel machine that s what caused their move to go and emulate exactly what Intel was doing and that s what enabled them to get a product at least into design as quickly as they did because they knew the approach They knew all of the clues and everything that we had decided to do Mazor Can turn a few more generalities if you don t mind and that is Ted and working for a memory company and Ted was confronted with a problem and his solution was Gee wonder if we can solve it by building memories Namely the ROM and the RAM in the 4000 series The other is we were presented with a real customer problem and when we were working on a CPU chip it had to solve the problem It had to be done to make that thing act like a desktop calculator So it wasn t creating a computer in the abstract but creating a computer to solve a particular customer s problem And we had a customer
3. House Okay so we got to the point where there is presentation made to the management and the decision was made to go with the Intel design think it would be interesting to hear from Ted and then from Shima san about their experiences from that point on Hoff We had the go ahead believe around October My group called Applications Research was not considered a design group The responsibility for chip design particularly MOS chip design fell under Les Vadasz So we turned the design over to Les and we did a little bit of you might say study on it to double check to make sure that we were reasonably close on our estimates with the help of some of the MOS design engineers But neither Stan nor had done chip design and didn t feel that a project of this importance should be given to amateurs that we would represent But it took Les quite a while to find somebody to do the next step in the project which was the detailed circuit and layout That didn t happen until the following April So we were in touch with Les during that time to see what was happening on the staffing In the meantime we did make a few changes That s when we decided to integrate the timing chip with the CPU and make it a true single chip CPU And we also started another project that came about from our success in selling this one That was the project that led to the 8008 But there was also application information that we continued to develop on the 4004 and o
4. but it was a couple months too late Otherwise they would ve been right to claim that they had done it Hoff There was another interesting aspect of it though They did not understand our interrupt capability Faggin Well neither do we because the interrupt capability on the 8008 was so late lt laughter gt Hoff It was a little well there were ways but the difference was that the TI chip did not change the restart instruction from a jam load to a call We made the restart a call so you could use that single byte instruction for the interrupt and they did not have it But at least in one of their patents they argue that you use the restart to implement an interrupt Which means you just wiped out the status of the processor Faggin The other thing which is important is that their chip size was exactly twice the chip size of the 8008 House Was it metal gate Faggin Metal gate So that s the only example that have that you can actually point and say Here is A B comparison This is twice the size of that and it s exactly the same function CHM Ref X4024 2007 2007 Computer History Museum Page 33 of 36 Oral History Panel of the Intel 4004 Microprocessor Mazor However if you want to look at it from a patent point of view in retrospect and if we could all learn from this and that is had we done the three instruction set computer we could ve patented the computer and we might have had patent rights th
5. Federico got it all on the chip Yeah understand there was some opposition inside even early on to this whole project also Tell us what Andy Grove and Les Vadasz attitude was towards the 4004 when you arrived Faggin Well think Andy s attitude was fairly old style actually remember was walking down the hall probably one or two weeks after was hired and said Andy this is a great thing It s a great project You know this thing is going to really change the way we do things And he looked at me with a face like saying What re you talking about lt laughter gt No but Vadasz was also of the same opinion it was a bit of a waste of time Not as strong as Andy But they were so busy doing the memory which was the main business of Intel that he had no time to dedicate to this side of the business that was considered kind of a poor orphan House A concern that this would distract from the basic business of making memories Faggin That s right with that the fact that they didn t hire the people in time And when joined had to resources and on and on and on So it clearly was not thought by the two of them as a line of business worthy of pursuit at that time House So marketing thinks that there s not enough big enough market and that their salesmen can t sell it and a number of these things And management s concerned that it might interfere with the memory business So who did you talk to o
6. House Yes go ahead Stan Mazor In September of 69 had been designing a floating point unit at Fairchild and had programmed at Fairchild the 1620 the IBM 1401 IBM 360 SDS 920 and programmed in about four languages Working on a high level language computer was a very exciting project Ted offered me a job to come to work on memory systems And from working on one of the most exciting computers in the world to go to work on memory systems was like Okay that s the end of your career but you know Intel s an exciting company So joined Ted with the idea that was going to be working on memories and not super excited about the idea But early on Ted said to me What would you think about a computer chip And he asked me to go home and think about it overnight Well I d been working on a giant computer and it happened to be stack oriented And so his suggestion for me to think about it was completely out of my mind frame And came back and mentioned a few ideas having to do with a stack based computer But he was an excellent manager because although at that time he had already conceived of this 4 bit processor he did not want to poison me in advance with his ideas So had very little to contribute on that day Then he disclosed the 4 bit computer that he had contemplated He had a separate timing chip and asked him Why is that timing chip separate Subsequently he thought Well gee maybe we could integ
7. by rubbing the tip on the metal we would simply break the metal And you know in cases you wanted to test what would happen with less signal So it was interesting how we were able to debug relatively quickly chips that were very complex where very few points were available House Was Shima in Japan now Faggin Oh yeah Shima had already been gone for a long time He had gone in the middle of October or so And the other thing is put a number of test points inside here on purpose so that could test different points where they had a little larger metal Little metal tabs so could go in and test internal nodes that otherwise would have been difficult to get at to help the debugging phase So basically at that point the only thing that made it not useful to Shima was that there was a problem with the program counter It would only count up to 16 and then it wouldn t ripple There was a piece missing So it couldn t really run a program Also it would ve been able to run a complete program because the other mistake was not as catastrophic but this one was a bad one House Sixteen location program was not Faggin Yeah was not a very good idea House And ran a calculator Mazor It ran a calculator too You know I ve met naive designers who didn t think much about testing One of the things that these guys had done is that during the operation of the CPU it put the contents of the internal bus out onto the pins They wer
8. devices were smaller about anywhere from 30 to 50 percent smaller and much more reliable So that technology was used on a commercial integrated circuit that was introduced in late 68 early 69 called the Fairchild 3708 That was the first silicon gate circuit that was commercially available and sold designed the circuit and actually build them in the R amp D fab So with that experience was the natural person to join Intel to lead the project of the 4004 in early 1970 House Hal Hal Feeney My name is Hal Feeney and was born and grew up in Davenport lowa went to the University of Notre Dame for my college training and it was at that point that was introduced to computers while working on my electrical engineering degree From Notre Dame went on to graduate school at Stanford and in the summers worked with Hughes Aircraft in Los Angeles California It was there that was first introduced to MOS technology At Hughes most of the activity was on military based systems or satellite based systems but we were doing some designs and started working with MOS technology prior to the time that it became available for large scale integration There were gates flip CHM Ref X4024 2007 2007 Computer History Museum Page 3 of 36 Oral History Panel of the Intel 4004 Microprocessor flops NAND gates NOR gates available in the 1965 timeframe We were using those to build up an initial digital version of an analog
9. gave us very good tracking of our customers because everyone had to be identified before we could ship any product to them But it raised one other obstacle and that is that we basically couldn t sell this product through distribution Everything had to be done directly with the factory So there were a lot of complications as we went along with that But the one piece that was missing and it came up very quickly out of some demonstration work that was done in Ted s group a board that demonstrated the 4000 series on a regular about foot square computer board with standard memory and the 4004 or the 4002s because they had I O with standard ROM but also again a feature that could use the I O off of those devices and then demonstrate them in a variety of different applications interfacing with the teletype and programming the EPROMs But will turn it over to Ted to discuss that a little bit more because that became the genesis ultimately for development systems that Intel would sell and would be a very strong business House remember at that point someone suggesting that since we were giving away so many manuals and selling so few chips that we sell the manuals and give away the chips Feeney Give away the chips yes Intel in the past typically had given away lots of free samples We made a decision at the time that the first microprocessors came out that we set prices on the microprocessor devices and did not give them away as f
10. go to Mendocino but there are lots and lots of wineries When went to Dr Noyce house was drunk lt laughter gt And stayed overnight Then came back to Silicon Valley Mazor Thank you Shima But I think Federico did an excellent job It was unbelievable to develop four kinds of chips within what was it Eight months Faggin A ha House So when did the last chip go to mask making Faggin It was October House October And then you had silicon Faggin And then the first silicon was in December pretty much on schedule CHM Ref X4024 2007 2007 Computer History Museum Page 16 of 36 Oral History Panel of the Intel 4004 Microprocessor House Now we re up to October of 1970 So tell us about October November December what happens next Faggin Well basically from that point on you have to go through the process of mask making and then the process of manufacturing the wafers So there was nothing happening for me until the wafers came out They came out toward the end of the year It was right after Christmas And so got my first wafers and was ready to go mean you know the tester was ready we got all the characterization sort test loops programs we re ready to test House Were those test programs developed after ruby out Faggin Yeah those were developed after ruby out and a lot of work was done by Hal and also some by Shima But mostly Shima worked on the production tape for the Pac
11. joined the company There was a different attitude moving ahead Faggin Yeah Hoff You know you have to realize think that we forget with computers being ubiquitous today they were much rarer in those days and not many people had experience with computers There were a few of us on the technical side who used them in design and so on but probably most of management never dealt with computers Feeney And we were very limited on how much timesharing computer power we could use at any given point in time along the way too even in the design of these computers Hoff So things have really changed since the announcement of that part But it was a few of us who really did believe that this thing had potential and fortunately we prevailed in getting Intel to announce it as a product Mazor have a datasheet here and it says on the datasheet that it s a micro programmable general purpose computer four bit parallel CPU with 45 instructions Now when we did an advertisement there were many people who did not believe that we had a computer They thought it was an arithmetic unit or something else And they really were offended but tell you when they looked at the datasheet it was compelling Hoff Another thing about it this was marketing s decision Data sheets at Intel at least up until this point had all the electrical information in the front And the application information in the back And it was decided for this pr
12. of forward looking ideas on applications since both of those applications came to market fruition Hoff have to say have very mixed feelings about marketing on a number of issues mean traveling with marketing people and seeing what they go through came away with the greatest respect but then on the other hand you say something like the scientific calculator which was discussing with the marketing people and they concluded Well the market might be for a 1 000 machines a year lt laughter gt And you say Well they re in the business They supposedly know Faggin It wasn t too long after the 4004 that the HP35 came out Hoff Exactly Faggin And so there were a bunch of other people thinking the same way Feeney Well as all of these things came together it became very much this convergence of technology and enabling functions smaller displays and being able to have the LED displays then liquid crystal displays And it s a matter of just having the right elements at the right time and think this is what Intel had with the 4000 series when it came to market They were the right elements to go in a system design and albeit not fast enough to meet all of the customers needs or meet the customers desires but they were able to be used in digital scales in cash registers in small control devices Traveling with Bob Noyce right after the announcement he was speaking from the point of view of
13. story of the actual development of the product Faggin Yes well in early 1970 was still working at Fairchild wanted to move from primarily a process development activity which is what did at Fairchild developing silicon gate technology for P channel N channel and CMOS wanted to actually design integrated circuits complex data circuits using silicon gate technology So contacted Les Vadasz early in 1970 because Intel was the only other company other than Fairchild that had silicon gate technology in those days Not much happened until March timeframe And then he was in a hurry for me to join the company At that time was having my wife was having a baby our first daughter remember that one day had to cancel an interview because my wife was in labor and Vadasz was mad that had to cancel an interview with him because we were having a baby lt laughter gt So anyway so eventually joined Intel in April the 3 1970 At that time met Stan Mazor Stan actually gave me the background of what work was done and he gave me a bunch of material that had accumulated specification of the CPU of the entire family a bunch of random circuit designs that were probably done in the design group that actually were not very useful as learned more about what was to be done My first task of course was to figure out what my project was but then the next day or the day after Shima was supposed to come from Japan and check o
14. subsequently I ve seen a copy of the February contract and it actually does talk about the potential for sales to other people but with the forgiveness of engineering charges should that take place But at that time don t think had seen that February contract But when marketing told us that they were going to be going to Japan because Busicom had asked for a reduction in the price of the chipset because the calculator business was becoming more competitive was one that went to marketing and said If you can t get any other concessions get the right to sell it to other people Now that was about May of 71 And they came back and said they had the rights then to sell it And believe there may have been a brief blurb put in think it was Datamation around May of 71 talking about the chips But it wasn t like an official announcement and it just kind of went away went to Noyce and to marketing people asking them to please get this thing available think it was probably early summer of 71 believe it was before Ed Gelbach joined the company But Bob Noyce said We have a tiger by the tail here but we are not ready to make a decision about announcing it as a product or adopting it And said Every time you put it off you ve made the decision not to announce And said Somebody else is gonna beat us to this if we don t get this out there House There were some reasons that were state
15. the market House Tells you how important it was for your wife to go to Italy Faggin Yeah Absolutely mean if we had been two or three months late TI would ve claimed that they were the inventor of the microprocessor right or wrong Hoff By the way did have a meeting with Gary Boone of TI at one point and Boone told me that TI did not start working on their version of the 8008 until after they heard about the Intel version They had the Intel target spec And they were working for Computer Terminal on a three chip set In other words the CHM Ref X4024 2007 2007 Computer History Museum Page 32 of 36 Oral History Panel of the Intel 4004 Microprocessor registers that they Computer Terminal originally approached us for was one of the three chips of the processor It wasn t until TI heard that Intel was going to do it all on one chip that they decided they would do a one chip CPU Faggin And that was about the time when started working on the 4004 Hoff Yeah Faggin In fact a little earlier because the first few months was working on the 4001 and 4002 and 4003 and therefore you know there was a delay from when joined the company Hoff This was actually probably before you joined Intel Faggin That s right Hoff Because that spec went out around January or February of 1970 Faggin It took them about a year and two months or so a year or three months to get to what they claim a working chip
16. then Barbara Mannis just helped about one month on the 4004 certainly want to acknowledge them They did a very good job House Any other names you want to mention Shima San Shima No told everything Hoff We might want to mention there was Norm Shanks that did quite a few of the sim boards and things like that which were very helpful in those early development tools House And we mention Ed Gelbach think Hank Smith had some roles CHM Ref X4024 2007 2007 Computer History Museum Page 35 of 36 Oral History Panel of the Intel 4004 Microprocessor Hoff Oh yes yes mean both Hank and Ed were very instrumental in getting this thing launched and putting together all of the support tools and so on that go with it The datasheets and user manuals and getting all that stuff published House Is there anything else anyone would like to add No Fine okay thank you for joining us all today Faggin Thank you Hoff Thank you Shima Thank you Mazor Thank you Feeney Thank you END OF INTERVIEW CHM Ref X4024 2007 2007 Computer History Museum Page 36 of 36
17. used relatively large numbers of pins for the time and our silicon gate process would have to be in ceramic packages because there were issues of the process and the impact of moisture on it So we had concerns that that packages used for some of the initial proposals of that LSI would make it difficult or impossible to meet the cost targets expressed some of these concerns to Bob Noyce and he said Well if you have any ideas to simplify the design go ahead and pursue them So starting with the design that Mr Shima and his team had started looking to see if could find some way to simplify it There were a couple of things that came to mind One was that the memory to be used in the original design was shift register memory which was dynamic and used six transistors per bit And Intel was starting to work on the three transistor dynamic RAM cel I that would be three transistors per bit that would be one way to cut down on the transistor count But the dynamic RAM seemed like it had another advantage in that it would allow you to simplify the control logic In other words if you work with relatively small chunks of information which you can do with a dynamic RAM but it s much harder with a shift register that suggested that maybe there would be a way to simplify the instruction set itself And one way to do that would be to go to a more primitive instruction set and make more use of sub routines and then essentially write routi
18. we did not think that Busicom would allow us to use the 4000 family for other than calculator applications right And so if you take that out you know the feeling was there would be very little left for the 4004 to be useful And that s what wanted to prove that in fact that was not the case House So how did Intel get the approval from Busicom to sell this to other customers Faggin Well at that point then there were a number of us pushing management to do that For one thing recommended Noyce to lower the price of the chips to Busicom and get a release from that That was just before knew that he was going to Japan to visit Busicom And that s after talked to Shima and knew that Busicom was in financial difficulty and they felt that they were paying too high a price for the chipset So that certainly was one of the elements that helped CHM Ref X4024 2007 2007 Computer History Museum Page 20 of 36 Oral History Panel of the Intel 4004 Microprocessor House What timeframe was that decision Faggin This was probably in the May June timeframe Around that time House Of 71 Faggin Of 71 yeah 71 Hoff think the impression of us not being interested in offering this as a product is wrong Actually we felt quite betrayed when marketing told us that they had given exclusive rights to Busicom for the set because we felt it had other applications even back in that October November 69 timeframe But
19. were going to go ahead with our approach at that time CHM Ref X4024 2007 2007 Computer History Museum Page 5 of 36 Oral History Panel of the Intel 4004 Microprocessor House Maybe we could ask Shima his reaction to the proposal that you made Shima About the end of August 1969 Ted Hoff came to our office He started to explain about his proposal And in his drawings there were three boxes and 4 bit input port for keyboard And those three boxes were 4 bit parallel arithmetic unit with accumulator and carry flag and 16 sets of 4 bit general purpose register units And four levels of address stack including a one program counter And Hoff s proposal was a 4 bit binary computer instead of a decimal computer felt that Hoff s proposal was good but if we accepted Hoff s proposal as it was we had to do the project over again from the beginning And also there were many difficulties at the time of evaluation of his proposal because information was too little But by mid September Mr Tanba of Busicom judged that Hoff s proposal looks more excellent than Busicom s proposal because the micro level instruction set architecture would also be the next generation s logical design method As he used to be a computer hardware engineer and in 1968 he had promoted the introduction of program logic method so in mid September Mr Tanba asked us to evaluate Hoff s proposal more deeply and carefully Mazor Dave can come back to that
20. 4 2007 2007 Computer History Museum Page 10 of 36 Oral History Panel of the Intel 4004 Microprocessor to complete the check of all circuit diagrams by April 25 Also Busicom requested that the production of 4002 RAM and 4003 Shift Register start at the end of July because the breadboard will be built at Busicom at the end of July Also Busicom requested Intel to substantially complete the production of the 4004 by the end of October This Busicom s proposed schedule was based on the letter from Intel on September 16 1969 And then on April 7 1970 visited Intel again for checking the logic schematic And when requested Faggin to submit the logic schematic he said he just joined Intel Then there is not any logic schematic at all but it was unexpected because considerable progress had been expected was so upset lt laughter gt Faggin Thank you Shima Thank you lt laughter gt Shima Yeah like Faggin said that next violently agreed and protested by a strong tongue lt laughter gt Then when Busicom s lawyer came to Intel and we requested to submit the development of a schedule it was found out the second engineer is scheduled to be hired in June for 4004 Since Busicom planned to develop the 4004 system hardware emulator for early stage functional verification at the end of July and variations Therefore Busicom proposed that in order to make progress on 4004 CPU development in schedule Shima will design 4004 logic i
21. And then I started the layout By the way had no layout people In fact we got a delay of at least two or three weeks to find somebody that would do the layout Finally they hire because in those days personnel will hire the person for you think you remember So they hire this layout draftsman called Rod Sayer from Lockheed He was a mechanical draftsman never seen an integrated circuit never even CHM Ref X4024 2007 2007 Computer History Museum Page 12 of 36 Oral History Panel of the Intel 4004 Microprocessor seen a transistor So basically had to do the layout of the 4001 a 2000 bit ROM had to draw by hand all of the building blocks and give them to him He would simply copy it into the Mylar So the 4001 layout was done the actual logic design took about a few weeks and the circuit design a few weeks developed a method where instead of doing circuit simulation that would take a lot of time had a method where used graphic design that would allow us to size the transistors without spending too much time also determined all the fundamental rules for the logic design for example no more than three levels of gates you know all the basic elements that would then constrain the design in a certain way so that you could actually do it effectively And so the 4001 layout took about a month and then of course came the checking Shima helped me throughout all the phases of checking in those early days And then we had t
22. ab assistant at the computer center at San Francisco State and left in 1963 to work at the University of California also as a programmer on a 1620 and that was the fanciest 1620 in the United States having many special features During that time had a chance to visit Stanford to meet with Ted Hoff who probably had the second fanciest 1620 in the United States and you could say we had a little bit of club of 1620 users In 1964 joined Fairchild and was at Fairchild Semiconductor in Mountain View from 1964 to 1969 For several of those years worked as a computer designer and logic design at Fairchild R amp D We had a project that was based upon Moore s Law and the fact that we thought hardware would become free and that was to develop a very large scale time sharing computer which also had decimal arithmetic in it And then left Fairchild to join Ted at Intel in September of 1964 House Ted Ted Hoff My name is Marcian Edward Hoff Junior but everybody calls me Ted was born in Rochester New York and did my undergraduate college work at Rensselaer Polytechnic Institute in Troy New York And while was going to college had a summer job at a company in Rochester General Railway Signal Company that gave me the opportunity to work with transistors and even a little bit of core memory came out to Stanford to do graduate work and got a PhD in 1962 and worked in the area of neural networks Some of that work inv
23. and punching paper tape in source code plugging in the four assembler ROMs on the board running his source code through the assembler and producing object code on the teletype unplugging his assembler programs programming from his paper tape into his EPROMs and plugging his EPROMs back in the board Now this sounds like a very messy deal but it did provide a mechanism for the customer with a teletype and this little board to create and develop and debug his own software CHM Ref X4024 2007 2007 Computer History Museum Page 29 of 36 Oral History Panel of the Intel 4004 Microprocessor Hoff It sounds like a horrendous procedure but anybody who had a PDP 8 and ran Fortran on it went through an ordeal that was an order of magnitude more difficult than that House Oh is that right Hoff That s correct Feeney Well think this is kind of the prototype of software distribution where you now go out and get software on a CD ROM and at that time we were using the 1702s as the vehicle for distribution But you know watching all of this evolve and watching our customers work with this was kind of interesting because it was after January of 72 that got involved in marketing and was doing the product marketing for all of our microprocessors for a number of years As the customers moved from this barebones type of development environment into finally a complete packaged computer with all of the software and all of the tools and eve
24. at would be important Whether it really is small or not and efficient or as good is a different story want to give kudos to my associates who did such a fine engineering job and pulled off this magnificent family that they did and also to my boss Ted have to tell you two more stories Ted came up with two ideas using the 4004 family at work One of them was a videogame and he had one of his engineers Glen Louie produce a game called Space War using the MCS 4 set on an ordinary TV set And so we were demonstrating a videogame Now one of the problems in the applications department you always have these sort of crazy ideas and you get to pursue them You don t get to go into that business and assure you when we showed a game on television to our management they said Who would ever want one of those lt laughter gt House That s a crazy idea Mazor Another thing that Ted came up with which was quite clever at one time we had an idea for a calculator chip a single chip calculator and then he had the idea of putting a ROM on that that was basically pushing the buttons of the calculator to do sine and cosine and higher level math functions as opposed to simple four function Again some of our slide rule carrying management couldn t really see that that one would have been an application for scientific calculators Now again it is the mania of an applications department to think of crazy ideas and Ted certainly had a lot
25. ause we were running programs out of read only memory Ted and had seen that on the IBM 1620 which had a sub routine return address think people subsequently have missed the significance of the stack the fact that Ted was working with three transistor dynamic memory cells with built in refresh Also it was kind a joke we had 16 four bit registers mean that s a lot of registers inside of a CPU but of course being only four bits wide would sort of make you chuckle And another important point with 16 pin packages is that there was very little room for communication control lines from the CPU to the ROM and RAM Another part of the architecture which is sometimes lost on people is the distributing of the op code decoding logic so that a ROM or RAM was looking at the same bus When the instruction was being fetched at the CPU the RAM could say Oh that s an instruction that I m going to perform rather than the CPU having to tell it the RAM that the RAM recognized it automatically And finally Mr Shima came back with at least four changes in the instruction set based upon the detail coding that he had done He mentioned the keyboard process and the jump instruction So we got it very far along into the project would say we were quite a bit influenced by the PDP 8 architecture because we had a lot of instructions that looked similar to the PDP 8 a few ideas from the IBM 1620 and then some of the finishing details based up
26. came out came out as a chip in October It was close to the time that Shima left for Japan And in the meantime they got a version of the tester ready so that could find out if this thing worked Mazor Can interject fora minute Intel was very small at that time and Federico was my officemate so after Shima s arrival the three of us were there together And Ted and are in the meantime working on other things the 8008 and so on But distinctly remember and you can help me with this that Elvia Federico s wife went to Italy in the summer probably with the new baby and Federico was working till midnight There was no wife at home and this was a project that was very absorbing So he put out an enormous effort during that time However being up till midnight he would sometimes come in a little bit late say 9 00 in the morning And at the time the rule at Intel was you get there at 8 00 a m So this was an interesting contradiction Faggin Yeah but anyway so the first chip to actually be completed was the 4001 As the layout of the 4001 was proceeding started the design of the 4003 The 4003 by the way used a new type of flip flop because it s a static shift register and is an idea that patented in Italy when worked for S T Micro back in 67 That basically uses fewer transistors to do a static flip flop And this was a simple chip It was a chip that the layout took a couple of weeks and it was done Sort of in
27. characterized the life of charge storage on the early EPROMs because it was thought they were just gonna be used on the bench But we found very quickly that customers were using them for production House remember putting EPROMs on the roof on the Bower s Avenue building Hoff Yeah House To see how long they d last out in the sun to see what d happen to them Hoff But we did the simulator boards with the idea that it would be a useful tool to help develop a microprocessor application and tried to make them duplicate what you would have when you went to the mask programmed 4001s and so on And provided the board such that you could use the EPROM for the read only memory part of it Initially the marketing people said we should just give the boards away And always argued didn t think that was a good idea because we would always be there looking for an excuse to stop giving away money argued that we should be able to sell the boards and even make a profit on them and it would be a lot less expensive for our customers to buy a board from us which we were producing in volume than to have to design it and build it from scratch themselves And on top of that this would be a tool to get our microprocessors into the marketplace because if the customer has got this hurdle that he s gotta design a prototype board himself that may be enough to stop him from using the microprocessor So it seemed like it worked out wel
28. chip need three more output ports or I need three input ports and an output So in a sense you re doing a custom mask for the ROM not only is the program configurable but also the port is configurable So the parts were fairly general purpose to the extent that the system was flexible both in terms that the amount of memory and the amount of IO Therefore we thought the applications would be quite broad FII just mention a couple of other quick ones while lm at it And that is having done a lot of software over the years knew that we would need an assembler for writing assembly language because that s how you wrote the MCS 4 And remember mentioning it to Gordon Moore who believed that an assembler is someone who works on the manufacturing line putting it together So was able to throw together an assembler on a weekend by using a macro assembler You may be familiar with high level macro assemblers You could create a macro for each suite of instructions The other minor point is that the mask that we produced for our customer had 256 bits and those bits were physical items on the mask and you actually had to generate a way of putting down those little bits So it was a minor support tool We wrote a program that took the ROM data that had to be stored and printed it out House And mapped it Mazor And mapped it literally so you could stick down the ones or zeroes that were applicable So that was some of the early s
29. computer a digital differential analyzer DDA And was working on the test equipment for the DDA when was first introduced to MOS From there did additional graduate work after Stanford at Notre Dame and then came to work for General Instrument Corporation in Los Angeles in 1968 GI was at that time the leader in MOS technology although it was a very very small business from a nationwide point of view GI was doing some custom work and some standard chips in MOS technology with probably in the ballpark of 500 to 1 000 gates on a chip at the time That was my introduction as a design engineer to doing individual chips did about three or four chips at Gl before leaving and joining Intel in 1970 In 1970 Intel was starting work on the 8008 Intel s first 8 bit microprocessor but due to some customer problems we slowed down the activity and spent time on several other projects one was working with the testing activity on the MCS 4 House Thank you Hal and thank you everyone So this all started out in Japan in the calculator business So if we can ask Shima San to tell us what stimulated the decision to develop a general purpose LSI family and how you came to select Intel to work with Shima In 1969 Busicom decided to develop a general purpose LSI family that would be used for not only a desktop calculator but also a business machine such as a billing machine cash register and a teller machine Also Busicom wanted to integra
30. correct House Intel and Busicom are working together now on finalizing the instruction set and making sure the application all happens So let s hear the view from the Intel side Mazor think it s interesting the synchronicity in your life because had worked on an interpreter in college and then had been working in decimal floating point on the IBM 1620 and then on the Symbol computer So here was again in my third project working for Ted and working with Mr Shima doing decimal floating point One of the problems had was that had a customer who would like to look a at the machine as he originally saw it in his flow charts as being a higher level machine but we had made a dumber simpler machine underneath So the idea of an interpreter came up Simply in about 19 bytes of 4004 code we could write an interpreter which would more or less present the view of Mr Shima s original machine that was capable of doing 16 digit add and 16 digit shift left which we implemented with sub routines and a 4004 code The linkage was a single byte because you could pick up the byte and use it as an address And one of the enhancements we made to the 4004 CPU was the ability to fetch data out of the read only memory and then to do an indirect jump That was what allowed us to do an interpreter for our customer But Ted had a number of innovations that want to call attention to The first was that the on chip stack was needed bec
31. cs for the planning of circuit design and layout design picked up the three A2 size of paper then considering the interface with each unit at first the general purpose register unit and address stack unit were drawn on the paper at right side And next the arithmetic unit the command control unit and the timing unit and the system bus control section were drawn on the paper at left side Then the instruction register decoder encoder and instruction execution control unit were drawn at the center of the paper As a result the 4004 layout became almost the same as the logic schematic It took about three months to design all of the 4004 logic including the logic simulation Fortunately there was only one mistake and this mistake was found out by both logic simulation and hardware emulator using different test vectors House Very good Faggin So we are in this process of developing these four chips one after another After the design of the 4002 was checked and while the layout was going on that s when Shima started doing the detailed logic design of the 4004 using a lot of the building blocks that were developed for the 4001 and4002 Particularly the memory and the memory refresh All those parts that Shima mentioned House So after the April when you started did Shima come immediately to live he just stayed Faggin Two days later two days after was hired yeah He never went back yeah CHM Ref X4024 2007 2007 Compute
32. cted a customer to be adding the RAM which was an oddball amount It was 320 bits but that s for another reason CHM Ref X4024 2007 2007 Computer History Museum Page 22 of 36 Oral History Panel of the Intel 4004 Microprocessor House That s for Mazor Yeah But in Shima s case believe he had four ROMs in his calculator and then later on he could do something like offer square root or some other function by adding a ROM So the maximum system that we provided direct support for was 16 ROMs and 16 RAMs So the system was very configurable Back on the point about minimum cost systems you could build a system with two chips or three chips down in the low end As you got into larger and larger systems think Ted s point would be at that point there might be more efficient ways of buying memory and perhaps therefore having a CPU Now another point that hasn t been stated but it was part of the architecture that think was worth mentioning and that is that when you got a RAM chip it also had a four bit output port So for every RAM you added to the system you have another four wires to attach to lights or switches or keyboard So in the Busicom case with two RAMs that gave them eight output signals But an additional feature was that with each ROM chip you got 256 bytes of program memory and a four bit IO port which was mask programmable That is the customer upon ordering the ROM says Well in this case on this ROM
33. d by people that were on the other side for not introducing this And maybe you might review what the opposition was Hoff Well there were several arguments that were made One was we d be competing with our customers for memory In other words our main customer base was the computer companies that made large mainframe and mini computers and they were going to be buying Intel memory and if we were in the computer business we d be seen as competing with them And so maybe they would go to other companies that did not have microprocessors in their portfolio think we tried to reassure that these computers were much smaller much lower in performance They weren t going to necessarily compete with a minicomputer If somebody needed a mini computer he s going to buy a minicomputer He s not gonna be able to use a mini computer to solve a microprocessor problem In other words the level of performance that the microprocessor had was significantly less The other thing was How will we ever support these And the statement heard was We ve got diode salesmen out there trying to sell memory chips How re we going to support computers And that s where Stan and would say Well we ll try to put together design aids application notes user manuals and things like that think we could probably hand hold maybe a dozen major customers and the rest of them we ll try to have enough information so they can make it out the
34. d the performance And the third problem was interpreter s programming method for application program That used the micro level instructions If the sub routine was simply used always two bytes of ROM are required for each micro level instruction And fourth problem was the real time control programming method in order to control multiple peripherals such as keyboard calculator display and printer It was terrible headache for Busicom The last problem was the instruction set itself In the first ten days of September 1969 the first batch of instruction set came out from Ted Hoff This instruction set was too primitive and only the decimal adjustment instruction set was added And Busicom team judged the primitive instruction set required a loss of size of memory and also it is impossible to control the multiple peripheral in the real time mode CHM Ref X4024 2007 2007 Computer History Museum Page 7 of 36 Oral History Panel of the Intel 4004 Microprocessor In September 1969 Stan Mazor joined the team The original proposal of Intel had been improved about two times step by step At first we generated generated a lot of desktop calculator s program using proposed instruction set Then performed a static operation dynamic operation for the proposed instruction set By this time Intel agreed to develop LSI family including RAM and ROM And also according to the first stage preparations if the improvement was done based on the req
35. dress update logic Next is the instruction register timing unit and the command control section and the system bus control section Since a three transistor DRAM cell was used for both the general purpose register and the address stack the address update logic was placed between address stack unit and internal system bus Then the contents of address stack were read out at Phase 1 time Then updated address was written back to the address stack at Phase 2 time And in order to simplify the testing added logic that the contents of accumulator and the carry flag are sent out through the 4 bit external system bus at the idling time Then its total number of transistor became 1 139 The remaining logics were a four bit arithmetic unit and an instruction decoder encoder and the logic of instruction execution controller The basic idea of four bit arithmetic logic was proposed by Ted Hoff And tried to use the instruction encoder for avoiding the random logic kept thinking about those logics day and night The number of transistors was 329 for arithmetic unit and 769 for instruction decoder encoder and execution controller Then finally with my counting method the total number of transistor of 4004 became 2 238 It had been approaching to the climax of the 4004 logical design There were only one circuit engineer and two master designers And started to make the logic schematic without the size of the transistor in order to use its schemati
36. e doing something more visible outside and out and essentially more testable think it s another good experience they had and it demonstrates good skill to do something like that House Yeah good forethought Faggin Yeah so then of course after debugging there was another change of the mask set By March shipped the first fully working 4004 to Shima And then think a week or two later Shima said It works Because he plugged it into his own calculator By that time he had done all the software but FII let him tell the story of that House So maybe you could tell us about receiving the first chips Shima Yeah My real goal was to develop the desktop calculator with printer It took about four months to develop the prototype of a desktop calculator with printer including the development of application programs and the 4001 ROM hardware emulator From the general idea of Busicom it the development lasted about two years and three months And in April 1971 finally the desktop calculator worked publicly was so excited And was quite happy The desktop calculator with a printer was CHM Ref X4024 2007 2007 Computer History Museum Page 18 of 36 Oral History Panel of the Intel 4004 Microprocessor constructed with one 4004 CPU and two 4002 RAM memory and four 4001 ROM memory and the three 4003 shift registers was able to add several value added functions by program And just after the development of the de
37. e months in Italy It was a hot summer for me So the layout of the 4004 started around the August timeframe with the memory part It was the easiest part in a sense because it was a regurgitation of what was done earlier And the key thing there was to what was keeping me awake at night was had dreams of not being able to finish the end of the chip because there would not be enough room to put what was required And had to erase and start all over again for a big chunk of the chip because remember we are doing this by hand So if you did make some mistakes you d have to erase and start all over again And of course given the tight schedule it was not an easy task to do because you have to determine this dimension first or the other Depending on which way you go you determine that dimension first and then you proceed down that way If at the end you bulge out then you got a problem right House It must be a rectangle Faggin That s right it must be a rectangle So in fact you can see that this was the part that was finished last Initially was hoping that could fit everything in a little less dimension here but then eventually was clear because also when started this the logic design wasn t finished so didn t know how much to plan for the control logic which is this part here So eventually it became clear that could fit everything in this dimension But had sleepless night until the very end when closed the lo
38. e prices And would you really want to you know save 80 when you re already committed to maybe 320 or so and throw away a factor of 10 in performance which is what the TTL would offer you over early microprocessors But if the application was one that didn t need much memory it would be less of a problem and that s why we looked at things like that We talked about things like an elevator controller All you have to do is store a little bit of data for each floor It s only a few bytes of information House Also guess there was an argument about the number of computers that were sold a year Hoff Well one of the arguments got from the marketing people was about the time was saying You should get the right to sell it said Look they only sell about 20 000 mini computers over each year And we re late to the market and you d be lucky to get 10 percent of it Thats 2 000 chips a year And they said It s just not worth the headaches of support and everything for a market of only 2 000 chips Mazor Can chime on here too Dave if you don t mind House Yep Mazor Just to reiterate a couple of things that might otherwise get lost If you needed a CPU in the system and if you re going to have some program you needed a ROM so the minimum system is really two chips and that s using the index registers of the CPU as a House Memory Mazor As memory which isn t much So typically we expe
39. ematic of general purpose LSI it was not accepted too well and we were told that there were too many kinds of LSI to be developed and most of them are CHM Ref X4024 2007 2007 Computer History Museum Page 4 of 36 Oral History Panel of the Intel 4004 Microprocessor too complicated and too many transistors We had trouble Therefore in order to solve the trouble showed the decimal computer s system architecture its instruction set and desktop calculator s program This idea was readily well accepted However Ted Hoff told me The decimal computer with micro level instruction set was difficult to implement and also there are not enough flexibilities House Okay so now let s turn to Ted Hoff and tell us about your experience of your first involvement with Busicom and their proposal Hoff had attended some of the meetings that led to the initial contract to develop the LSI that were held in April of 1969 So had an idea of what some of the cost targets were but had not seen any specific description of the LSI and did not see that until Mr Shima and his team of engineers came over in June In looking at the design that they were bringing to us especially because Intel was relatively small and there were relatively few people who were capable of doing MOS LSI design started to have some concerns that this may be more complex than we had expected And another issue if remember correctly some of the packages
40. er than calculators that could use the 4004 Now at that time the way remember it both Mazor and Hoff were skeptical that applications for a general purpose control with the 4000 family was going to be good So took it upon myself to see if could use it in a practical application The opportunity came when had to build the production tester for the 4004 where used the 4004 as the controller for the tester And so developed a program used the 1702 the EPROMs that were available then Basically did what a customer would have to do to use it and learning how that could happen And at the end of that process which worked out quite well for me really tried to convince everybody that it was a good product to go out in the marketplace And at that point think the management of Intel started getting convinced that it could be a practical product to go on the market And so that was the beginning of a movement toward marketing the 4004 Of course the 8008 was in the wings That was from the very beginning considered good enough for other applications in the marketplace but initially not the 4000 family That s the way remember it Feeney But the 8008 was oriented toward data terminals character manipulation and whatever The 4000 series was in the control function part of it And think at that time management really didn t understand the difference between control computers and data processing computers Faggin But also
41. ere was fortunate enough to co design a small experimental computer and build it had four technicians working for me We built this machine several hundred printed circuit boards 4k core memory with 12 bit words That was an invaluable experience for me understanding what was going on in the computer field That computer of course used transistors in 1960 1961 Integrated circuits were not really available yet And used germanium transistors which were very very slow And yet the computer was fully functional and it was an interesting background for the 4004 later on then went to university where studied solid state physics with a doctorate in physics And then worked for what today is called S T Micro in those days it was called SGS Fairchild And developed there process technology for MOS integrated circuits which was the first time that S T Micro had that technology also designed a couple circuits using that technology Then in early 68 came to this country working for Fairchild Semiconductor in the R amp D laboratory because of the connection between SGS Fairchild and Fairchild There was fortunate to co develop was the principal developer and the project leader silicon gate technology This was the first technology using self aligned gates that really revolutionized MOS integrated circuits Because with that technology one could have three to five times the speed of the older metal gate technology The
42. ific Western tester So got the wafers at 6 00 PM or so so people were just leaving was actually happy that nobody was around the last thing you want is for people to ask Does it work Does it work So yeah too much pressure right And so loaded the first wafer in the probe station We had a probe station with probes that would go onto the pads and then you come down with the probes into the pads like this and then you send signals and you test the signals Generally in the early debugging stage you use a oscilloscope to see if you get what you expect And so went down to the first die and in the scope there is like floating nothing happens Oh okay well you know So go to a die floating nothing happens Well okay Another die Same thing And then said Well it s a bad wafer So took another wafer put it underneath go down same thing Now by this time was trembling was sweating lt laughter gt said How could have fucked up so bad You know fortunately the 4001 worked first time 4003 worked first time 4002 had a couple of minor problems So could not have screwed up so bad okay So took the wafer under the microscope look under the microscope and lo and behold they forgot the layer for the buried contact So a good 50 percent of the gates were floating and there was no signal so obviously the whole chip didn t work So it basically wasted three weeks at that time which wa
43. ions Prolog was a company that was a big user of those boards for control applications where they could interface simply with existing standard memories And then later on came up with the idea of the 4040 which was an improvement over the 4004 because it doubled your registers and you had a real interrupt instead of a flimsy test pin and a bunch of additions That product was intended to interface with standard memories and to give a mid life kicker to the 4004 which was clearly going to die some time later because a new generation 8080 and the N channel technology was coming up House And the 4040 came out when Faggin The 4040 came out early 74 and then it probably was in production until probably 78 79 something like that House And that sort of gave birth to the 8048 CHM Ref X4024 2007 2007 Computer History Museum Page 30 of 36 Oral History Panel of the Intel 4004 Microprocessor Faggin That eventually was superseded by the 8048 which was introduced in 77 78 Mazor So just let me reiterate 8048 is a single chip with ROM RAM and CPU on one chip and more or less followed in the controller applications that the 4004 carved out And so think Intel s product line then divides into these microcontrollers that contain programmable memory and microprocessors that do not Hoff 4004 gave birth to the embedded controller and 8008 to the data processor Faggin Yeah that is clearly a distincti
44. l and then we turned it over to the marketing department and they just took off with it and actually developed a full blown development system business which guess for some years was quite a money producer for Intel House Produced more money than the microprocessor business did for a lot of years That was true And understand that out of that came the first sim board that had to program EPROMs So how did the EPROM programmer evolve CHM Ref X4024 2007 2007 Computer History Museum Page 28 of 36 Oral History Panel of the Intel 4004 Microprocessor Hoff Well we had two EPROM programmers One was just a board that plugged into a sim module But we had also been charged with building what we called an EPROM burner And the first one we did we worked with an outside company that did wire wrap and we gave them a logic diagram and they had some problems We had a hard time getting that thing to work And about that time the microprocessors became available and we said What are we fooling around with this random logic for Let s use the microprocessor to build the EPROM burner And we did that House Was that done in your group Hoff That was done in my group think we called it the 7600 if remember the number of that And we had one funny problem when we were debugging it We had a paper tape reader on it and for some reason it was reading the paper tape wrong and we were convinced we had a bug in our program And i
45. lable to us were very slow paper tape readers that would run on a teletype machine As a result we took advantage of generating programs on the teletype machine and then loading them very quickly into a very deep memory using the high speed optical tape reader And the best price memory we had available at the time was Intel memory It was the 3101 bipolar memory So we built a rack of 3101 bipolar memory put the appropriate interface with it loaded it up with the tape and then we could set it up and time it and do a comparison with the operation of the processor We would send instructions to the processor and then also be able to have the outputs that we we re expecting on the tape So we send the instruction and have a comparison with the bus outputs that we expected and go through step by step instruction by instruction We could either set up a program and do something useful from a program point of view or the initial time of testing each device go through instruction by instruction by instruction and insure that each and every element of the program worked House It was largely for engineering debug CHM Ref X4024 2007 2007 Computer History Museum Page 19 of 36 Oral History Panel of the Intel 4004 Microprocessor Feeney This is the engineering debug and characterization part of it This is not the production testing part that would go in for any kind of production But was highly motivated to do this because knew that w
46. leting evaluations and also Busicom in Tokyo has been evaluating new instruction set for the scientific calculators And finally October 21 Busicom sent the data to Intel for adopting Intel s proposal But it took about two more months to finalize product specification of LSI And several instructions were added in order to get the higher performance and reduce the program size Such as like a subtract carry from accumulator or decimal subtractions a code converter for the keyboard scanning and clear for both accumulator and carry And a compliment carry and no operations Next with my best recollections in order to avoid misunderstanding made functional specification of instruction set in English adding lots of drawing and the equations but it might be in April in 1970 On December 16 in 1969 the last meeting was held with Intel Intel s skeptic to add two new instruction set CHM Ref X4024 2007 2007 Computer History Museum Page 8 of 36 Oral History Panel of the Intel 4004 Microprocessor such as a subtract carry from the accumulator and a quota convergence for keyboard Just felt that Intel didn t realize how it is important to keep the problems from our side to be small That is my comment for the Hoff s ideas Then returned home to Japan at the end of December House Okay so it s at that point in time when the specification is done and you re waiting for Vadasz to staff the project Is that right Hoff That s
47. n the progress of the work At that time didn t know they were late Nobody told me they were late And so Stan and went to the airport to greet Shima Shima was tired but still wanted to come to work immediately He said I m here to check said Oh oh okay Great Just come over So Shima and Stan and walk into the office and Shima says Where is logic of CPU And said Well here s what got from Mazor yesterday This is all the material that have Take a look that s all know at this point think it was an hour later Shima comes back and says This is just idea This no good Now late Now late You bad lt laughter gt House Okay maybe Shima wants to comment Shima After returned to Japan took about two months to develop the program of desktop calculator with a new printer And made sure that there was no problem in the functional specification of the 4004 The formal agreement was concluded on February 6 in 1970 It said that Intel shall exercise its reasonable best effort within its capabilities And also Busicom to join in product development as requested by Intel And on March 20 1970 Busicom sent the desktop to Dr Noyce with a diagram of each LSI with explanations and a disk of instruction set And also its letter described The following Busicom requested that Intel substantially complete circuit diagram by Shima s visit on April 7 In order CHM Ref X402
48. n the 8008 That is looking at these devices and seeing what we were going to need to do to be able to use them in actual applications House So Shima san tell us your experience after the decision to go with the Intel design Shima After the Ted Hoff idea came out it took about four months to define all of the detailed product specifications And want to talk about what problems were there and how we solved the problems together Busicom team started to discuss with Ted Hoff about his proposal more deeply And by the way wish to understand the difference between the idea and product In other words there is a big difference between the research and development If the idea was kept of these it will not be successful Always the initial idea should be improved getting feedback from the applications And the Busicom team found out five problems to be solved The first problem was the lack of concept of LSI family The first Hoff proposal was to develop on the binary processor using a standard RAM and ROM It required a system bus interface to the LSI Also still in his proposal there were timing LSI printer control LSI and printer buffer LSI The requirement of Busicom with the highest priority was to build the system with only LSI product The second problem was the decimal data calculation method In the beginning Intel recommended us to use a method of look up table for decimal calculations However it required lots of ROM and lowere
49. n the project Faggin Moore didn t seem to have much of an understanding or interest in that At least not to me House Moore Faggin Gordon Moore yeah And then really it wasn t until Gelbach joined the company that Gelbach having lived in a company like TI where there was a fair amount of business with random logic and chips of that sort calculator chips and so on You had much appreciation of that part of the business And so Gelbach was my entr e into the marketing side and tried to convince him to go to market with the 4004 And so he was much more receptive as a matter of fact House When was the actual introduction Faggin It was in November CHM Ref X4024 2007 2007 Computer History Museum Page 24 of 36 Oral History Panel of the Intel 4004 Microprocessor House November 71 Faggin Yeah that s right But the decision was really done in early summer kind of June July timeframe at least the preliminary decision that we re going to go out and then probably the final decision was made in September would say timeframe But at that point all the plan of the introduction was done and laid out and all the documentation was being prepared so at that point it was irreversible So it was really September Is that what you remember roughly September was the irreversible decision Hoff Could be don t even know when that decision was made It always seemed to me that things changed once Ed Gelbach
50. ncluding the logic simulation and CPU test prototype generation and layout checking House So what happened next Faggin So what happened next is that had now this task where am essentially six months late the day after start have four chips to develop an angry customer and basically Intel had no experience whatsoever in designing random logic circuits particularly with silicon gate which is a new technology that requires a new methodology a new way of doing it Intel was a memory company They had done memories before and those designs are very different from random logic was on my own Vadasz my boss was completely absorbed by the memory activity And the application group it was not even their job to help me so basically was on my own Of course tried to do my best to make the customer happy felt that even if had nothing to do with it Intel was late and the customer was right in complaining Eventually was able to calm him down and say Shima if you help me then we can probably reduce the delay in the schedule And if not do whatever you want but there is nothing can do to make up the time Particularly since the original schedule was already wrong because in the original schedule the layout time of the 4004 at four or five weeks which was the layout time of the memory Therefore there was no way that it could be done It tells you that Intel was not a random logic company They didn t understa
51. nd what it took to do a random logic circuit Now had to give a little bit of background here because it s important to understand the state of the technology in those days Basically companies like Fairchild or Rockwell Semiconductor or GI that were in the random logic design with MOS technology had an entire methodology They had computer programs to do the logic simulation the circuit design simulation They had building blocks that were pre characterized They had examples of layouts how you do different types of circuit They had the experience of how to determine how long it takes to do design also how many transistors it would take to do things and so on All of that CHM Ref X4024 2007 2007 Computer History Museum Page 11 of 36 Oral History Panel of the Intel 4004 Microprocessor In addition a company like that would have for example testers for characterizing the product They would have production testers for those products All that infrastructure There was none of that at Intel None of that In addition they were attempting to do a two phase design without a bootstrap load which is ano no You cannot do that And of course in those days everybody knew that a bootstrap load cannot be done with silicon gate technology because you cannot make a capacitor with silicon gate technology the same way you do it with metal gate And so the entire approach was not possible You could do fully static designs which would be fine e
52. nes that would simulate the more complicated chip set started pursuing some of those ideas in the July and August timeframe of 1969 Another idea that came to me was that the initial design used mostly binary coded decimal arithmetic It seemed that with a primitive instruction set it would be possible to do binary arithmetic and have a primitive instruction that would convert a binary result back to valid binary coded decimal So it would be possible to make a machine that was both binary and decimal arithmetic in its capabilities started to put together a proposed instruction set and checked to see that it could perform a number of the different calculator functions In fact it also looked like it would be possible to simulate some of the operations that were done by the other chips like the keyboard controller chip and the printer control chip Around the first of September Stan Mazor joined my group and together we put together a crude form of target spec Around the middle of September our marketing department made a formal proposal back to the Japanese calculator company Busicom suggesting that they might want to consider looking at this other design And if remember correctly we had a meeting somewhere around October of 69 Mr Shima and his team presented their approach and Stan and presented our approach to the management of Busicom and they at that point said they really did like the Intel approach So that looked like we
53. o create the Rubylith As you probably remember you put the Mylar with the entire design on a cutting table and then you put a layer of Rubylith which is a fine layer of red Mylar over a base of transparent Mylar and then on a backlit cutting table you cut traces and then you peel off the unwanted material And that is a very laborious process prone to errors Not only is it prone to errors when you do it but also when you transport the Mylar Sometimes a piece of red Mylar would fall off When you had checked and it was fine but then in transportation it got corrupted and then you find out when you have a problem in the mask So it was during the design of the 4001 and 4002 that it was apparent to me that the less translations we do during the design the less errors we re going to make It was particularly obvious as got involved with these high complexity circuits So decided that instead of doing a separate logic design and then a circuit design to do logic and circuit design together in a single sheet But also with the notion of the layout so that to the extent possible you try to put the lines and the transistor placement in this sheet as close as you can guess to what it will be on the final layout Obviously you had to do the overall planning of the chip ahead of time so that you know roughly where the various blocks are It was during this period that refined the methodology of doing these type of designs So the 4001 then
54. oduct the application information would be in the front and the electrical information would be in the back lt laughter gt It was moving from components to systems Feeney Well Ted don t think it was quite that dramatic at the time The whole issue was to have a datasheet And on that datasheet you ve got the features at the top you ve got the pin out on there and everything else There was a very distinct problem with the MCS 4 because you needed to have four pin outs So there was no way to get all four pin outs on the front page so the very first thing violated the Intel tradition CHM Ref X4024 2007 2007 Computer History Museum Page 25 of 36 Oral History Panel of the Intel 4004 Microprocessor Hoff do remember somebody in marketing telling me This was a milestone for Intel going away from the traditional datasheet House From the traditional memory datasheet Feeney mean it s a matter of how these things are done as an Intel style and this was part of the mold to break of Intel style Yes we do shift registers we do memories Processors are a lot different and then by the way there s a user s manual book with it call in and get the book that tells you really how to use this thing Mazor There re other anecdotal stories I d like to share one A fellow called me up from the telephone company He said he had the datasheet He was going to quit his job and do something with this chipse
55. oftware support Hoff There s one other point too When we heard that the 4004 was not likely to be available to other customers besides Busicom but we had the 8008 in the line and the 8008 was supposedly going to be available to any and all customers assuming that Intel decided to go into business Stan and designed another four bit processor with the idea that if we did not get rights to the 4004 we would have a low end processor And this one actually could run out of standard memory In other words it could directly talk to an EPROM or a standard read write memory We called it the 4005 We Intel never produced it but as part of a sale of technology to MIL Microsystems International Limited of Canada we transferred the design of the 4005 to MIL of Canada And believe they actually did produce the processor House It was never introduced as a product that we know of Hoff think it was by MIL of Canada but Intel never produced it and sold it CHM Ref X4024 2007 2007 Computer History Museum Page 23 of 36 Oral History Panel of the Intel 4004 Microprocessor Mazor Not only that one of the things that we were concerned about was that on the 4004 we were too ambitious so this 4005 was a greatly simplified four bit computer even in comparison to the 4004 For example it had no index registers at all and no stack and so it only had like two pointers so it was a very simplified four bit computer House In fact
56. olved using the IBM 1620 computer That was really my first computer experience Later we replaced it with an IBM 1130 and continued working after got my degree as a research associate And then was approached by Bob Noyce who told me he was starting a company had been interested in where integrated circuits were going and had actually done some work at Stanford at least looking at integrated circuits and so felt that the next area was memory When Noyce interviewed me he asked me where though the company should go and said Memory and it turned out that s where Intel was going So left Stanford and joined Intel in 1968 My job there was called Manager of Applications Research It was mostly to look at how to use the semiconductor memories how to help customers use them and to also help define what the products should be So that s how got started Intel was going to do this development of what were considered proprietary products that would take some time to design in to persuade computer companies to switch from the core memories they used at the time and so Intel was open to the idea of doing some custom work Our first custom project was to be done for a company we knew as Electro Technical Industries And that s where first met Mr Shima House Thank you Ted Shima San could you give us an update on your background before you started working on this product CHM Ref X4024 2007 2007 Computer Histo
57. on almost all the applications of the 4004 early on were embedded applications Yeah maybe not a single chip embedded application but an embedded application Hoff That s one of the things too that the media today still seems to be clueless about embedded control They always think of the desktop computer as computers but the degree to which embedded control is used everywhere is something that seems to be lost on them Faggin Well it s at least a factor of 20 1 in microcontrollers versus microprocessors in consumption At least 20 1 Hoff At least Feeney Even inside of the computers that are sold today Feeney Yeah the keyboard controller the Ethernet controllers the Faggin The touchpad controller it goes down the line There re probably 20 microcontrollers in a single PC Mazor If could also add something about what was going on in the competition Maybe my friends here can add something as well As we came out with the 4004 there were companies in the semiconductor business who thought we were making a big mistake because we were coming out with a brand new architecture and a brand new instruction set and that meant that the customer base had no existing software that they could utilize So RCA as an example decided to make a PDP 8 compatible chip think they called COSMAC However the fact that their class of machine stores the return address for subroutine calls in main memory made it only usef
58. on Mr Shima s experience House And that brings us up to the point at which it became time to develop the chip Ted do you have any more comments CHM Ref X4024 2007 2007 Computer History Museum Page 9 of 36 Oral History Panel of the Intel 4004 Microprocessor Hoff Well no think there were just so many things that seemed to work in our favor There were just little things that seemed to fall into place To give an idea we used a three step transmission cycle to send out the address to the program ROMs and the size of the ROM was 256 bytes which takes eight bits of address So the first two nibbles that go out go to all of the ROMs and each one can be making access to its data and then the third nibble that comes along tells which ROM gets to go on the bus That way we didn t have to sit around and wait for the ROMs In other words by sending the bytes low order middle order and high order in that step cycle the ROMs were ready with data and whichever one was selected then put its data on the bus So there are all sorts of little details like that that just fell into place House So now understand Some time goes by and Intel engineering is busy doing other things and not performing according to the contract in terms of delivering these products until Les Vadasz hired you Federico to come in and take it over Maybe you could tell us your experience when you first arrived and how you got involved and then go on into the
59. op CHM Ref X4024 2007 2007 Computer History Museum Page 15 of 36 Oral History Panel of the Intel 4004 Microprocessor around So the 4004 layout took about three months At the peak we had three people working on it and three layout draftsmen House Do you recall their names Faggin Yeah there was Rod Sayer Julie Hendrix and Barbara Mannis Barbara Mannis only worked for about a month and the other two were there for the entire duration of the project The challenge in the layout was fundamentally to do it fast enough You know other than of course closing the loop But then we had to tape out meaning what we call tape out now In those days it was the ruby out House Ruby out yes Faggin Ruby out was basically right after or around the time that Shima went back to Japan Shima helped me also with the checking of the ruby It was a daunting task because it s a complex chip and the Mylar wouldn t fit on the cutting table so it was in two pieces and all that So eventually we got into mask making Mazor want to ask Shima a question In his book he shows a picture of a big station wagon that he owned in the USA and also a picture of him at a company barbecue could you comment on that Shima Yeah when came to the States the second time didn t have a TV didn t have a car Therefore Dr Noyce told me You can use my wife s car but car is located at Mendocino And then one guy pick up me to
60. ould use exactly the same tool in exactly the same way to go through and do the 8008 when we had working chips with that Federico was talking about going home at 3 00 in the morning and one thing we learned about that is that there s very little Silicon Valley traffic and you can get home very quickly at 3 00 in the morning lt laughter gt With no problems at all Yeah and we had a lot of experience doing that But setting up having the test equipment being able to go through and do almost anything we wanted to with these chips at that point it encouraged us also that we can play with these things we can program them and thinking of a variety of things that customers could do if we ever had customers outside of Busicom to use these devices This was the job that we had to go through from March up until the following November when the product was released to customers outside of Busicom Faggin Well from the time got the project was very excited about the opportunity to develop the 4000 family because saw and Ted and Mazor saw the enormous potential of this type of device However early on this project was an exclusive project for Busicom so it could not be sold to anybody else always felt that the 4004 was good for a number of control applications because even in the calculator or in the billing machine you have a number of peripherals that you need to control So felt that there would be all kinds of other applications oth
61. quired to meet the economic side of the business So my first task was to understand what was to do found that there were a couple of mistakes in the architecture fixed those Then there was another problem There was a power on clear Basically you had to have the chip coming up in a predetermined state Nobody knew how to do that It was important because it would require another pin and there were no extra pins because we were using all 16 pins So then came up with another invention In fact we got a patent Intel got a patent on that power on clear circuit a specialized flip flop that always is assured to come up in a certain way So then immediately started working put a schedule together that to my best of my knowledge at that time could give an idea to the customer of when we could have devices The best that could do is to have first silicon of the 4004 in December of 1970 But couldn t do any better than that Even that would have required the chip to work first time which is a tall order for a complex chip like this Then decided to start by doing the 4001 first then the 4003 then the 4002 then the 4004 so that it would be a warm up The 4001 and 4002 had a fair amount of logic so that would require me to develop the methodology in a more gradual way more organic way And could have all the pieces together for the time when we were required to do the 4004 And so did the design of the 4001 Shima checked it
62. r History Museum Page 14 of 36 Oral History Panel of the Intel 4004 Microprocessor Shima was captured Faggin Through October yeah through October he was House Captured Faggin Yeah that s right six months So of course Shima didn t know how to design integrated circuits and he didn t know transistors or any of that but he learned very rapidly Shima knew a little bit Faggin Yeah he knew a little bit like a transistors a switch Shima That was enough It was digital Faggin Yeah except for the RAM Yeah so Shima in the first few months also learned a lot through the design of the 4001 4002 4003 We learned a tremendous amount about how to take some of the stuff that did on those chips and put it into the logic design of the 4004 And of course it was a tremendous to have Shima to actually help in that phase of the project because that time was the peak of activity had the 4001 in mask making and the 4003 in rubylith cutting and the 4002 in layout and starting the 4004 layout all at the same time only me to go around So Shima also worked very hard He would probably go home typically at 8 00 at night right But typically when he would go home was still there And at that time my wife as Mazor mentioned she left in June with our three year old daughter and went to see the grandparents was very happy to see her gone because that actually gave me a lot of time She stayed over thre
63. rate it So was a bit of a sounding board seeing it for the first time and asking questions about it We were located at 365 Middlefield It was an extremely small building and my office had three desks in it My officemates were Mr Shima and Mr Takayama was living with those gentlemen who were our customers and was able to see their excellent flow charts They had the design of a floating point unit which is done to a large extent in program code also understood that they had a range of calculators that they were interested in including display and printer One of the machines even had a card reader could see that they had some very grand goals for their project and they would ask very difficult questions One that remember well shows how expert Mr Shima was on his product He would ask me a question like How are we going to print in red when the negative result is computed Of course he knew that that was a requirement of the calculator while we were working on the other hand on a very low level 4 bit computer So was in the interesting position of trying to balance the needs that they had thought about quite a bit down to the details of printing and our CHM Ref X4024 2007 2007 Computer History Museum Page 6 of 36 Oral History Panel of the Intel 4004 Microprocessor implementation So my general role during that time was liaison in working with Ted and with Mr Shima in finding the solutions to these problems
64. rd such as signal from printer The definition of conditional branch instruction was modified for judging the test input And thirdly return from the sub routine instruction was modified to store return address to be used for the conditional branch micro level instruction And fourthly new LSI was added for the keyboard scanning and also the print buffer and also its LSI combat 10 bit serial input data to the 10 bit parallel output data Fifthly to secure the flexibility number of status register in ROM LSI was enlarged to be four With these added instruction set the modification and the new LSI it became possible to control many peripheral in their time by program And next in order to delete the decoder MSI for chip selection of RAM and ROM a new instruction to designate the command line for chip selection was added In the original proposal there was only store instruction for data transfer from accumulator to general purpose register So load instruction and exchange instruction were added deleting the store instruction If the number of general purpose register is not enough the exchange is quite powerful instruction set Also it was very difficult decision to choose shift instruction or rotator instruction Then we choose rotator instruction for the high flexibility and the reduction of application program size And October 14 again Mr Graham sent the data for asking to accept Intel s proposal Still needed one more week for comp
65. re on their own Now kind of figured if this thing was successful the feedback will take care of itself CHM Ref X4024 2007 2007 Computer History Museum Page 21 of 36 Oral History Panel of the Intel 4004 Microprocessor House understand there was an argument also Hoff There was a concern that too many people who had used mini computers would come to use one of the microprocessors and it s offering probably an order of magnitude less performance and that they might be disappointed That was the biggest concern that they would say I d rather spend 80 more or 100 more and buy TTL and build a processor rather than buy your 20 chip and not be able to do what need Fortunately that didn t happen That was probably the only thing where we had any reservation about offering it as a product House As understand there was a concern that the memory cross would be so large that reducing the cost of the processor wouldn t be that important Hoff Well for some problems that would be the case because our target when we started Intel or actually some time after we started Intel was to get the cost of memory to a penny a bit by 1972 You have to figure this is happening even prior to 1972 to make this decision whether to offer this as a product And it doesn t take very much memory figure what d ya take 3000 characters to store a printed page Well that s 300 worth of memory approximately at thos
66. ree samples as a way 1 of qualifying our customers and 2 demonstrating that there is significant value in having a processor on a chip Faggin And number 3 paying for the manuals lt laughter gt Feeney Very true Mazor Just quickly a background point and that is when Intel got started the primary memory in computers was core memory And core memory was non volatile That is if you turned off power at night came back in the morning and turned on power presumably your program was there intact Now in reality anybody who worked in the computer center probably didn t really rely on that fact But one of the objections that we had in doing semiconductor memory is we had to say right up front that it was volatile And more or less we had to overcome the objection of the customer which is Well wait a minute you know if you turn off power you re gonna lose all the data While don t think this is really a serious issue but we did have to confront it So one of the products that was outlined that we needed at Intel was a non volatile memory And it was understood in the early 70s that the nitride process might give you a way of treating the gate in such a way that it would remember information when power was off So another officemate that had when Faggin moved on was Dov Frohman Bentchkowsky And he was given the project of creating a nitride memory Well to do that he had to study the gate silicon dioxide interface and
67. ry Museum Page 2 of 36 Oral History Panel of the Intel 4004 Microprocessor Masatoshi Shima My name is Masatoshi Shima was born in Shizuoka in Japan graduated from the Department of Chemistry at Tohoku University in Japan in 1967 In April 1967 joined the Busicom Corporation started to work on a program at the computer division and learned several programming languages such as Fortran COBOL and assembly language on Mitsubishi s computers MELCOM 3100 And six months later in October 1967 moved to the desktop calculator division and developed a desktop calculator that was implemented by hardwired TTL logic The design of a printed circuit board became quite useful for the design of the 4004 In those days all companies in business of desktop calculator grew up very rapidly in Japan and Busicom started to look for the new logic design methodology and methods that would be suitable for OEM business Then in November 1968 developed new desktop calculator with printer In that development introduced the program logic method using the decimal computer and read only memory for program storage The data types were n digits decimal data and five bits binary data House Thank you Federico Federico Faggin Yes my name is Federico Faggin was born in Vicenza Italy where went to a technical high school So by the time was 18 graduated And my first job was at Olivetti in the R amp D laboratory near Milan And th
68. rything that they needed it expanded our market But the customers really wanted to get in work very closely with the product and on the original SIM4 01 boards that were available they could get in and get to quite a variety of different test points on the board and see exactly what their systems would be doing and have a very good tool for emulating those systems And then with all of the random logic that was there it became fairly clear that if that logic could be replaced by something integrating the random logic we could even make the customer experience better and that goes to Federico where he carried things through further with the 4008 and the 4009 Faggin Yeah it was obvious that people wanted a microprocessor but standard memories also Also as time went by memories kept on getting better faster and cheaper and more bits and so on And we would have to continue to redesign the 4001 4000 bits and then 8000 bits and so on so it would ve been an expensive proposition to keep that family growing at the state of the art level so it was clear that we could have a way to build a system with the 4004 that was cheaper by developing a couple of chips that would interface the 4004 to extended memories and those chips designed very quickly because they were relatively simple But came up with the idea that was the 4008 and 4009 that proved a boon A number of customers were selling boards single board computers for control applicat
69. s not welcome because of course Shima was there waiting for his chips in Japan right And but you know there was nothing that could be done at that point House So you had to go back Faggin And so we had to go back Finally we got the new batch of units out in January some time in the second half of January or so And same thing they came out at 6 30 in the evening and now this time put down the probes and Aha Signals lt laughter gt So it looks good and run a little test loop and it passes and so was very excited at that point Everything that was testing seemed to be working kept on going on until 3 30 in the morning 3 30 4 00 in the morning and check all the major blocks at that point And then went home was exhausted And so at that point it was clear that we had a 4004 because even if there were problems which there were a couple of problems they would ve been easily fixable because all the major things were working And so then the following week or two debugged the whole thing and found those two problems What was interesting is that in those days we had little probes that have very sharp tips We would go inside the circuit itself and probe in metal lines House So you had no glass layer on top CHM Ref X4024 2007 2007 Computer History Museum Page 17 of 36 Oral History Panel of the Intel 4004 Microprocessor Faggin That s right There would be no glass layer Sometimes just
70. sktop calculator Busicom developed the Cash register the Billing Machine and the Teller machine for NCR Japan And let me put my comment The microprocessor replaced the hard wired logic by program After the microprocessor was developed the functional specification of the system was decided by the capacity of memory and the performance of the microprocessor And was quite happy to work with Intel Thank you Faggin Thank you House And so Faggin And so then after that of course came the task of transferring to production all these parts House Is this when Hal Feeney got involved Faggin Hal Feeney got involved toward the September House Hal maybe you could tell your experience of the 4004 project Feeney Well as Federico said got involved in it at the time that we were kind of delaying or running the 8008 at a low level and got involved helping him almost any way that could But the main project was a very deep memory that would allow us to functionally test some of this stuff Because again we re sending an instruction to the processor we re getting information back from the processor we re comparing and so forth And we had no tools that would allow us to do this or allow us to have any kind of along program Now at the same time in Ted Hoff s group Tor Lund was working on a PROM programmer On the PROM programmer he had a paper tape reader The only other paper tape readers that we had avai
71. t Thats how dramatic it was Shima But the Japanese market system makers were looking for the microprocessor type of device For example a cash register To make a cash register it require like 200 to 500 TTL packages A huge power supply Thing like 4000 line came out they tried to look for the microprocessor quite deeply House It was taken more seriously in Japan Shima Uhm hm House So now if we can turn to building out the product line You ve got some chips and you ve decided to make it a product but it s really a computer and out of this came the genesis of a systems level business at Intel and simulator boards and this sort of thing House Hal maybe you want to talk about the marketing efforts to launch this new concept Feeney Okay well got involved with the marketing side of it kind of in two fold way was finishing up the 8008 in engineering in the fall or fourth quarter of 1971 and it was November 15 that we announced the 4004 or the MCS 4 that was our evolution in the way of thinking of the family We went from being a company that would just announce chips to announcing a complete family of products for the very first time And on announcing the MCS 4 we called it the alternative to random logic And with an alternative to random logic it was a matter of replacing TTL that had been used heavily for gates flip flops other slightly higher levels of integration and be able to replace those TTL de
72. t turned out the driver to the paper tape reader had a snubber diode on the coil and that was slowing the paper tape reader down But we actually wrote a simulator of the 4004 to check it out and the simulator said it should work And the hardware wasn t working House What did the simulator run on Hoff Well the simulator ran on another machine House Like a DEC PDP 10 or something Hoff Yeah So that was one case And then we had contact with some of the outside suppliers of EPROM and PROM programmers and we pointed out the advantages of microprocessors And about that time Intel switched from think it was the 1702 to the 1702A or it was some change like that that had a major impact on the programming algorithm And we redid our burner in like one day with just you know House Because it was programmable Hoff It s all programmable you know And we pointed that out to the people making the EPROM burners they should be using microprocessors in theirs too and they took that to heart Mazor The genesis of this product line was that we had the SIM4 01 board with EPROMs on it that allowed you to emulate your final program But our customer was confronted with the issue of writing a symbolic program and assembling it So a little bit further in that gestation we actually had an assembler written in 4004 code and stored in EPROMs It ran on a SIM4 01 board So the customer s process went typing up on a teletype
73. te a lot of function on LSI Therefore Busicom choose Intel as a co developer because Intel had already developed an advanced high density and high performance silicon gate MOS process technology The provisional agreement was concluded on April 28 in 1969 However Busicom didn t disclose to Intel about the application of a business machine because it was the confidential matter between Busicom and NCR Japan House How many people were on this team in Japan and who was leading the team Tell us a little bit about that Shima There were three engineers from Japan the project manager Mr Masuda also the senior design engineer Mr Takayama and was the youngest engineer House How did you get involved in the project and what was your role in the Busicom team Shima In Busicom my main role was to develop the decimal computer s system architecture instruction set architecture and application program In the beginning planned to develop nine kinds of LSI but later it changed to eight Those were two LSI for decimal computer ROM LSI for program storage and shift register LSI for data storage Then keyboard display controller LSI printer controller LSI and print buffer LSI And then we visited Intel on June 20 1969 House So when you presented your proposal to Intel Intel saw some problems or had some other ideas Can you tell us how you tried to address that problem Shima When Busicom showed to the Intel all of sch
74. the spare time And then after this was done the design of the 4002 occurred Shima checked all along The 4002 layout started while we were doing the rublith cutting on the 4001 So basically we were doing pipeline designs designing four chips in parallel slightly offset one from the other From April to December we designed four chips of which three were LSI in basically eight months This was not the easiest thing to do given that a typical chip like a 4004 would take normally a year by itself see that Shima wants to say some words CHM Ref X4024 2007 2007 Computer History Museum Page 13 of 36 Oral History Panel of the Intel 4004 Microprocessor House Tell us about the challenges that you faced during this time Shima Shima The most difficult logical design goal was to use a small number of transistors At first a detailed function block diagram of 4004 was drawn before its detail logical design And also the interfaces between the function units were clarified Because it s work inference both chip size and performance acquired heavily repeated it several times until was convinced So as not to waste one transistor started the detail logic design There were four units and two sections such that lots of transistors are needed but design is relatively easy Those are the general purpose register unit with two bit refresh counter the address stack unit with two bit stack pointer two bit refresh counter and four bit ad
75. tion for the 4005 Now it turns out eventually we competed in the marketplace with a very successful product called the Rockwell PPS 4 If you studied the PPS 4 will claim you will find an awful lot of similarities between that and the 4005 and it turns out that particular calculator manufacturer did a lot of business with Rockwell when Rockwell was new in the calculator business think Shima can also tell a story about the 4004 going to a Japanese manufacturer and perhaps getting picked up Shima One day a 4005 stick went to the Sharp And Sharp gave that stick to NEC Then NEC started to develop the four bit microprocessor and it was quite successful House So what did NEC call that Shima ViewCom 4 House ViewCom 4 okay Mazor So my point is there may be more origins of the microcomputer history here than just the Intel ones House Some legitimate and some bastard Faggin Yeah but there s not a question that even in 1970 71 it was a race mean really mean if you look at TI for example we helped them it looks like It looked like but still they were developing an eight bit processor which was essentially the 8008 And they came out in April May timeframe with that product We heard heard from Vic Poor that it never worked But TI claims that it did work In any event whether it worked or not it was later than the 4004 by one or two months That tells you how much execution was important to being first in
76. to do that he had to get not much metal on the chip and actually it was during those experiments that he called me over one day and said Look at this and he showed me on a curve CHM Ref X4024 2007 2007 Computer History Museum Page 27 of 36 Oral History Panel of the Intel 4004 Microprocessor tracer what we call a logic One and he blew on the chip and it became a logic Zero So he had come through with a mechanism which we called the floating gate for storing charge And in this case in a non passivated chip the moisture in his breath could cause the current to leak out Interestingly enough management didn t believe in it too much and gave him quite a bit of latitude But in the early days of Intel we developed this new product which became the EPROM And EPROM became an essential synergistic product with a microprocessor because our customers would want to try out programs and the way they would do that was with erasable or reprogrammable EPROMs House remember we used to sell the EPROMs as only an engineering tool And you d obviously go to masked ROM when you went to production Of course because there were always software changes the customers actually went into production with EPROM which was a nice surprise for us because we got a lot more money for EPROMs Hoff was gonna just add that because the initial view was that they were only a development tool often wondered does anybody know to what extent they
77. uirement from Busicom Intel proposal is acceptable On September 13 Mr Masuda the project manager returned to Japan with result of faster stage operations On September 16 Mr Graham who was marketing manager of Intel sent the data to Busicom to request the adoption of Intel s proposal Also he strongly requested to use the program logic method for peripheral control Also its letter said that chief development takes only six months with Intel s idea instead of one year with Busicom s proposal This proposed schedule became a big problem when visited Intel in next year for logic checking And in order to achieve a high level of performance by using the low performance processor and a small size of ROM memory proposed a new instruction set and function modifications and a new LSI as well as getting the good advice from Intel And felt that Intel tried to avoid having an abundant instruction set to keep the number of transistor as low as possible At that time Intel set the target of number of transistor to be about 2 000 And still believe 4004 project was not able to succeed without feedback from the system users and application program My story may be a little bit long House It s okay Shima At first in order to make the interpreters program the 8 bit immediate data loading instruction with register in direct addressing mode was added And secondly the test interpreter being was added to examine the extent of external wo
78. ul for a system with read write memory and not with ROM So it promptly failed in the controller market Fairchild which flipped over what we did decided that what the world really needed was a bit serial machine and they came out with a 16 bit processor but it was bit serial which we had rejected So it s interesting that we created some ripples in the marketplace and competition responded can tell two anecdotal stories not with a high degree of certainty but with regard to the 4005 my office mate again was someone from MIL Ken Au who was their engineer My role was to translate those specifications and get them stored in the logic design CHM Ref X4024 2007 2007 Computer History Museum Page 31 of 36 Oral History Panel of the Intel 4004 Microprocessor Well when Faggin succeeded with the 4004 we abandoned our efforts MIL went back and did that chip But in the end we did compete in the marketplace with the 4005 can remember a personal situation as an applications engineer where I m confronted with a competitive situation and it was 4004 you know our little son versus the 4005 our little daughter lt laughter gt and had to argue why one was better than the other in this case the Intel product The other is that was called upon to go to a calculator manufacturer and to give them a specification But we couldn t give them the 4004 specification This was Victor Comptometer in Chicago And did give them a specifica
79. vices with programmable semiconductor devices Now we had to go through several things One we had to present it as a component and we talked about that with the datasheet that Stan showed Beyond that we had to do a user s manual that would pull together all of the instructions and tell people how to program the device and beyond that it was a matter of showing them and giving them tools to use Well at the time of the introduction we basically had the Alternative marketing brochure we had the datasheet we had a user s manual we printed lots and lots of users manuals and gave them out to everybody that called in and went into numerous reprintings of the user s manuals because it seemed CHM Ref X4024 2007 2007 Computer History Museum Page 26 of 36 Oral History Panel of the Intel 4004 Microprocessor the customers had just an insatiable appetite to get information on these new devices So everything was set up to bring this microprocessor product family into the market The one obstacle or speed bump we had with this was that because we had permission from Busicom to sell the product to other customers Intel sold the MCS 4 with the caveat that the product could not be used in any calculator applications Therefore before anybody could get a sample of the product they had to sign off on a disclaimer that the product would not be used in calculator applications And for each and every customer we had to do this But it also
80. who knew very well what he needed and that gave us the opportunity He didn t change his mind from day to day So we had a customer who knew what he needed and was able to see how to apply our products to it House And so there were a number of other people that were involved in the program Obviously we can t include everyone but one person we invited here today was not able to make it for a health reason was Tom Rowe who worked in the process side So he s a silicon process guy working on perfecting the various processes that were used and making sure that they yield and we can actually make these products And possibly there were others that should be mentioned Faggin Well in my group Paul Matrovich was my technician and he really helped in all phases of testing building the testers You know we built a team of electronics for the tester that Hal was mentioning earlier so that we could also set levels and detect some of the more analog kind of measurements on the chip And Charlie Corbin another technician that came in toward the very end of the project Then Yung Fang was an engineer that came in also toward the end of the project and helped doing the detail characterization taking measurements and making sure that the product was working under all kind of conditions temperature and voltage and so on And think mentioned the three draftsmen Rod Sayer was there for all the chips Julie Hendrix did the 4002 and the 4004 And
81. xcept that they take two to three times more transistors House Space and power Faggin That s right more than conventional dynamic design Two phase designs would require a bootstrap load Four phase designs which actually were the ones that were used for random logic circuits in those days because they had the best speed power product did not require bootstraping Although in metal gate you could do a bootstrap That s why Rockwell for example had four phase designs and that s how they got their calculator business So Intel with a two phase design which is an effective design methodology did not have the bootstrap load Fortunately had developed the bootstrap load at Fairchild just before coming over and had tested it out to see that it actually worked That required an invention but it was really more an understanding of the physics of how devices work because in a bootstrap load what you need to do is to always have a virtual junction instead of a physical junction The other thing that immediately was clear to me was that what was needed was the direct contact between buried poly silicon and the junction That was another invention that made at Fairchild back in 68 as a matter of fact and Vadasz knew about that That invention would allow you to have roughly half the chip size that it would take with metal gate to do the same function That invention was essential to get the chip size of the 4004 to the level that was re

Oral History Panel on the Development and Promotion of the Intel

Contents

Download Pdf Manuals

Related Search

Related Contents