Home

AN1722/D:SDRAM System Design Using the MPC106

image

Contents

1. vev8r9LL8gN 91 8XIN8 Y9XN91 8 82 v8y9An9L90S er fs 91 91 91 SL 6 e 91 92 60 sv 9 lsz 92 Sv 9 ALA V2p9lPOLISaN 9 9 v9XW8 GWv9 viv9An8eoQs cl 0 Sc 0 Sc 01 JOL JOL 0 Sc 0 Sc v Gv 0 Sy v 0 l l 8 91188 8 8 8 yOXW8 vy8v9An8g9QS 61 iz a fer fer ler fse ze ze e 69 69 69 69 69 69 AL ALALA 3228211189001 91 8XNZ Y9XWv gWee Gc8v9Anvoas 6L fer fer jer e e 69 69 69 69 69 6 A A A A YZZ8LLLL8AN 91 Y9XNv Wee Oc8v9AnvoQs cL jo e D Je 6 6 6 lo le o je 0 fse le ze ALA 3ec8 11188N 8 Y XNZ gW9i Gz8v9Anzoas zt jo Je o je e 6 je lo js o ze zc o ue vue ue ALA YZZ8ZLLL8gN 9 Y XNZ GOL Ozsp9ANzoas cl 0 eb 0 el 6 6 6 0 SL 0 ke 1 1 D Ic ile d E T 0091 11188 Y 9 Y9XNL awe v eLvgAntoqgs nsunj yor ys ss ws US ss yo ys ss Ws svo svy jueuoduio ed ZHIN Set ZHIN 001 ZHIN 8 S9 anoa x19 339 19 yueg uoneziuebio ezis ued suz ouae 5 2 us senjeA 4d eoueyoede yids 49019 Wvuas eed 9iger For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Organization and Loading Factors THIS PAGE INTENTIONALLY LEFT BLANK SDRAM System Design Using the
2. 81 vXW ZXWv SINCE 000 6 jor o se o se jor jor jor je lsz 60 60 oe og sp 8 lop A A 91 Y9XNv GWcE 000 9 e fe e e je jor o poe o os le ier fe o oe pe lss lss o ss lss SS 1 1 6 6 exe zXW2 8 91 O002AZ SAH e fe e e b le 6 jor o se o se or jor jor o 0 se lsr ser o St 1 L AIA 8 L v9XWZ 8 91 0002AY9SAH e fe je e 6 jor o oz o ye 9 ir c lo oe je joe lo oe joc oe 1 L 2 SA S 9IXNL 0001AZZSAH e u le e je je jo jo st jo sg jor jor jor jo jo lo ge se je 0 62 s Sc 1 ai Lia v 91XNL 9 8 8 000LAY9SAH Sueueig e k 9 e k e y e e e le poc 0 69 o 9 joe oe 06 or sr or jsp se se 109 09 se 196 A SA A SA 81 vXN9L ZLXN9L 8901821 LYOZILSYLSNNWA E z p le e z e e joe jo ss 0 lss 60 60 se 0 or or lop 06 06 99 99 06 06 A AJ AIA 94 vXN9L 9 9 8 1811 17029159954 e e o e l e ue e e e s se fse lse se or or jor jor loe loe 99 ss 06 06 AJA 91 exe z Y XNIL 8 4811 LYEZILSIISWWA E z p le E z e e se 0 jov 0 jr sc sc o 0 jspr 9 09 09 o 109 109 09 L L A SA 6 8XW8 4 ZLXW8 GINS lvezeSv SINI E z p le e z le t e e se 0 o j lsz sc sc o 0 lop 5 55 ise ise se S L L ALA 8 OLX z 9 8 GWv9 LYYZBSIISNNWA E k 9 e k e wW e e e le loz 0 se 0 loz loz oz o lo or 99 es o gs ss sS 1 L ALA 8 8XW8 9X8 GW
3. 972 66 100 2 1 5ns 4 5ns SDMA12 SDMA11 SDMA2 SDBAO SDBA1 SDMAO SDMA1 DOMB 0 7 RASO_CSO RAS1_CS1 MPC106 WE KE DP 0 7 D 0 63 VA f DP 0 7 MPC60x DP 0 7 D 0 63 Figure 25 MPC106 SDRAM Interconnection This connection is sufficient to control a single one bank or two bank JEDEC DIMM Additional CS lines remain for up to three additional modules loading restrictions permitted of course 1 13 Conclusion It is possible to design a high speed SDRAM based system if careful attention is paid to the timing parameters parasitic conditions and the layout SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Revision History 1 14 Revision History Table 25 shows the revision history of this document Table 25 Revision History Revision 1 Updates incorporated Nontechnical reformatting SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Bibliography Appendix A Bibliography Table 26 lists useful reference documentation Table 26 Reference Documentation Martin Graham Description Author Document MPC106 PCI Bridge Memory Controller User s Manual Freescale MPC106UM AD MPC106 Revision 4 0 Supplement and User s Manual Errata Freescale MPC106UMAD A
4. L je 6 SI jv amp ve e jez jez 6 Ov jov Ov O 88 88 jvG pS 88 88 9 81 8X2 gWce OP6EZyNELWAl je e le 6 1 loe loe joe oe je jo joz jr jov jr jr log log os jos jog 08 9 v9XWv GgWee Of6r9rNELNGI e le je e 6 o o por o loz loz jz jr jor jr jv 98 09 05 08 98 81 8xWz Z XWv OO6 ZYNELWGI e le e e p le je o o vv o jr jer ger jv jr jr jv log log v v log 06 91 8XWc V9XWNv GINZE OO6v9vNELWGI e le je e l je jor o oc o joe jer ger jer jov lsz jor cs cs 0 65 jes les A SA 6 89XWc ZLXNZ IN91 OP6EZ2NELWGI e le 8 e 6 lo pe o ye jet gr 91 sz lop lsz jov 09 los o jos js los ALA 8 8XWc Y XNZ GIN9L Of6P9ZNELWal e u fe je je l f je 6 0 jo jo jo 91 jor ist 92 jov 0 joe o oe joe joe LIA v9XWI G8 ONGPOINELWal Wal er 60 60 joe joe 1 91 lsz 60 60 joe loe se s6 jos los 96 196 A A YZY8v9LL8gN 84 8XN8 gINBZl Y8ZLAN9LOAS zt o o joe jer el fe 61 le ez 0 js los o og js los A SA vev8vollggN 6 8X8 Z XN8 GSWv9 Y8ZLAN8IAS 61 jee lee ler ler ce jee pL L v nv VL UL A SA YZZ8LLLLBAN 84 8XNZ ZLXNY OcSz2AnvoQS Zi o o vi vi le SL joz 91 joe jov jov 0 jov 9 lor A SA YZZ8LLLL8gN 6 zXWe 8 9 v zgzzAnzoas 6L 66 90 98 90 jot St 61 jez 90 Sc 60 68 98 0 68 98
5. NORMAL INPUT A m OFFSET m CLK Tsus E OFFSET INPUT LLL LLC Tonos m Tons OFFSET 1 Tuis out LLL KO Figure 19 MPC106 Offset Clock Timing Diagram Figure 19 has zero margin since Tsus starts at the end of Top the important feature to note is that Toys has been delayed to that time at a minimum additional delay is acceptable to increase margins Essentially the purpose of the clock offset is to give the SDRAM extra setup time by taking it from the output hold time This is possible in that as noted before the TOF adds to the apparent hold time so subtracting some violates no timing requirements The only concern that arises is the clock to output timing of the SDRAM data bus which because it is typically lightly loaded and feeds directly into a registered transceiver can usually be compensated for but be sure to read the section on the data bus A revised margin table is shown in Table 19 in which a nominal amount of clock offset has been added a sufficient amount to give a small positive margin in all cases SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Timing Analysis Measurement TOF Margin Clock offset 1 7 2 8 0 3 2 1 0 3 2 1 0 3 2 1 0 3 2 1 The margins are the same in all cases over 66 MHz because the clock offset was chosen that would giv
6. MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Organization and Loading Factors THIS PAGE INTENTIONALLY LEFT BLANK SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc How to Reach Us Home Page www freescale com E mail support freescale com USA Europe or Locations Not Listed Freescale Semiconductor Technical Information Center CH370 1300 N Alma School Road Chandler Arizona 85224 1 800 521 6274 or 1 480 768 2130 support freescale com Europe Middle East and Africa Freescale Halbleiter Deutschland GmbH Technical Information Center Schatzbogen 7 81829 Muenchen Germany 44 1296 380 456 English 46 8 52200080 English 49 89 92103 559 German 33 1 69 35 48 48 French support freescale com Japan Freescale Semiconductor Japan Ltd Headquarters ARCO Tower 15F 1 8 1 Shimo Meguro Meguro ku Tokyo 153 0064 Japan 0120 191014 or 81 3 5437 9125 support japan freescale com Asia Pacific Freescale Semiconductor Hong Kong Ltd Technical Information Center 2 Dai King Street Tai Po Industrial Estate Tai Po N T Hong Kong 800 2666 8080 support asia freescale com For Literature Requests Only Freescale Semiconductor Literature Distribution Center P O Box 5405 Denver Colorado 80217 1 800 441 2447 or 303 675 2140 Fax 303 675 2150 LDCForFree
7. The most common means of terminating high speed digital circuits are parallel series and AC termination SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Termination 1 7 2 Parallel Termination Parallel termination also called Thevenin or split termination is an ideal means of terminating high speed signals Although the total power dissipation is higher than other methods it is not as severe as may be believed With the reduction of voltage levels from 5 to 3 3 volts power dissipation has also fallen in an inverse square proportion Additionally the shorter rise time and ability to daisy chain signals with minimal reflections will be attractive to reach 100 MHz operation Figure 9 Parallel Termination Calculating the parameters for a parallel termination are slightly more involved because several constraints must be met e RIII R2 should equal Zo jy must not be exceeded y must not be exceeded The first calculation is easy the latter ones are expressed by examining the voltage drop across R1 or R2 when the output driver is at the active low or high state respectively Complicating matters is that the output level is not guaranteed to be VCC 3 3V or ground but instead could be at Voy and The equations to verify then are VCC Von Von RI R2 gt loh and VCC V Vai R1 R27 ol These are simultaneous equa
8. e perm ees Ex em e E uu qua a Es e pem emen SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Example System Table 24 MPC106 SDRAM Interface Connections continued SDRAM 168 Pin JEDEC 39 amp 126 depends 4 bank 2 bank operation SDBAO SDMA 1 12 P1 T16 R16 P15 M 38 121 37 P16 N16 M15 M16 120 36 119 35 L15 K15 K16 J16 118 34 117 33 Unlike the address lines the CS 0 7 and DQMB 0 7 lines are not reversed The reason for this is that as long as the data byte lanes to which the DQMB signal correspond DQMBO for D 0 7 DOMBI for D 8 15 etc the order is unimportant even within the byte Often in order to ease layout entire byte lanes and the corresponding control line may be swapped The address lines cannot be arbitrarily rearranged for two reasons first the least significant bits are important for correctly processing burst transfers second the most significant bits are unconnected in some of the smaller DIMM modules 1 12 1 The Big Picture With the previous elements in mind Figure 25 shows how the MPC106 would be attached to an SDRAM DIMM module SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Conclusion 1 0ns
9. is a measure of the ability to propagate an electric charge through a particular medium in this case a rectangular slab of copper overlying a large ground plane or power plane For thin traces where the width TW is greater than or equal to the height above the ground plane TH the permittivity of the trace is defined by the following equation r 17 r r 1 12 TH1 02 TW3 ETHIN 5 5 p m 0 04 1 zx Note that if the trace width is equal to the height above the ground plane the final term disappears Additionally at these small dimensions the permittivity of the surrounding air must also be considered so the effective permittivity of the trace becomes TT Er 1 ITW Once the permittivity of the trace is known the propagation delay can be computed EEFF THIN ATRACE 84 72 eae The effective trace width is needed to calculate the other measurements 1 25 TT 2 TH TWggg TW L eJ This in turn allows calculation of the characteristic impedance of the trace The formula for this is 120 7 ZWIDE mA CC ee PPP ae Es TW TW ER 1 393 0 667 n E 1 444 TH The impedance is important to know later when signal termination is discussed When PCBs are purchased the impedance shown above will be the ideal goal for the manufacturers to attain SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com
10. the TOF by scaling by the trace length Figure 8 shows the propagation delay for the previous set of loading categories and trace length Heavy 24 l Moderate Light Propagation Delay ns 1 2 3 4 5 6 Trace Length in Figure 8 Time of Flight 1 6 1 Point to Point Timing If the system consists of only one SDRAM module or memory then the TOF is complete as is however if there are multiple SDRAM modules connected serially then the TOF will differ for each device with devices closest to the MPC106 getting data earlier and those at the end last SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Table 10 Propagation Delay and TOF Device Distance Delay TOF Units MPC106 0 0 0 ns SDRAM 1 3 2 5 ns SDRAM 2 4 2 8 ns SDRAM 3 5 5 A 3 1 ns Termination For most systems it will be sufficient to insure that the final device in the chain receives data with the specified timing budget 1 7 Termination It is virtually impossible to design a high speed digital system without paying attention to signal termination and an MPC106 based SDRAM system is certainly no different At the speeds discussed in this document pay careful attention to terminating all of the control signals otherwise the signals may be corrupted at the destination the SDRAM leading to
11. there is no reason less or more margin could not be used provided all other SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc The Data Path constraints are not violated The choice for a margin amount must be an educated guess on the part of the system designer If the target system is expected to have any of the following problem sources then a larger margin is needed e Electrically noisy High levels of crosstalk dense PCBs e Poorly regulated noisy power supply e Insufficient ground return paths e Excessive jitter from clock source e Bombarded with cosmic rays For several of the desktop reference platforms from Freescale 300 ps is sufficient since more aggressive consideration was paid to the problem source list Some of those approaches e Thevenin termination on critical traces for cleaner signals e Extensive pre and post route simulation to minimize crosstalk e Qualified clock generator for minimal jitter e Good by pass placement for ground return paths If these cannot be achieved whether due to cost or time limitations it may be necessary to increase the timing margins to insure reliability 1 10 The Data Path Up to this point the data bus has been largely ignored and without the data path the SDRAM becomes just a big expensive write only memory It s time now to analyze the data path Fortunately just a
12. to route signals point to point 1 7 4 AC Termination If the signal can be assured of DC balance in which the net assumes that the logic high and logic low state are roughly equally the AC termination may be used as shown in Figure 11 To Figure 11 AC Termination AC termination is generally only suitable in an MPC106 based system for the SDRAM clocks all other signals exhibit significant amounts of time in either the high or low state The attractiveness of AC termination arises from the ability of the periodic transitions of the signal to charge the capacitor to an average voltage level of VCC 2 This in turn cuts the power dissipation to es P AVG Zo In the example of 65 board traces the power dissipation is 41 mW less than half that of the parallel terminator AC terminators are fairly simple to design the impedance of the termination resistor Z equals that of the board impedance The capacitor Cj is chosen such that the RC time constant of the two allows the voltage to decay during the clock interval For a 10 ns clock a suitable C would be around 150 pF AC terminations share the benefits of parallel terminators to which they are closely related such as point to point routing while eliminating the undesirable power wasted 1 7 5 Summary Table 12 summarizes the salient features of each type of termination The actual values for rise time and power dissipation vary depending upon the components
13. 0 PCICHIP 1 n 42 5 42 5 The manner in which Table 23 is used is to first layout the clock traces paying attention to usual routing restrictions on clock traces but not to length Try to keep the traces on the reference point as short as possible then produce a trace length report on the above signals The longest of the PCICARDS 1 n clock traces then becomes the reference point or those of PCICHIP if no slots are used First adjust the PCICARDS nets so they are all of equal length REF These nets become the new reference point for the remainder of the nets Next a 2 5 allowance is added to all clock traces except the slots to allow for the clock trace length on the PCI motherboard Obviously it pays to keep the PCICARDS clock traces as short as possible to eliminate extra routing added to all other traces Next a small amount of trace is removed from socketed devices such as the PowerPC processor cache module but not for direct attach BGA devices SDRAM DIMM modules if used and perhaps L2 cache modules At this point measuring the clock skew directly between any two device pins with an oscilloscope would reveal 0 0 ns skew ideally The remainder of Table 23 applies in turn the compensation factors for the MPC106 errata and the clock offset for high speed operation In some cases a lot of trace length must be added 1 11 2 Control Routing Once the clock signals have been routed the control si
14. 06 For More Information On This Product Go to www freescale com 8 Mbytes 16 Mbytes 64 Mbytes 64 Mbytes 32 Mbytes 128 Mbytes 16 Mbytes 64 Mbytes 32 Mbytes 128 Mbytes 32 Mbytes 32 Mbytes 32 Mbytes 4 1Mx16 8 2Mx8 8 8Mx8 8 4Mx16 16 2Mx8 16 8Mx8 18 2Mx8 18 8Mx8 18 2Mx8 18 4Mx8 18 2Mx8 Freescale Semiconductor Inc Clocks Table 13 Common SDRAM Clock Architectures CLK2 c The ideal clocking system is shown Figure 13 where there is one clock driver for each SDRAM clock pin in the system as well as ones for the MPC106 CPU memory data bus transceivers L2 L3 cache etc SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Clocks Low Skew Clock Driver Figure 13 Ideal Clock Architecture Not shown in Figure 13 are the clocks for the CPU cache MPC106 PCI slots and devices etc This architecture usually becomes prohibitively expensive when there are many SDRAM modules memories because there may not be enough clock signals to control the remainder of the system An alternate possibility is a distributed clock generation system as shown in Figure 14 Here a single system clock driver signal drives a remote PLL based clock driver These secondary clock chips synchronize themselves to the input source allowing the outputs to maintain 0 ns skew with relation to the ma
15. AM Component Selection In order to properly connect to the SDRAM allowances must be made for the capacitive loading of the signals for each SDRAM device or module present Each manufacturer of SDRAM memory devices has slight variations of I O drivers which causes a somewhat wide range of capacitances the design will have to compensate for Most desktop designs use the standardization and flexibility afforded by the JEDEC 168 pin DIMM SDRAM memory module or the small outline DIMM modules which is very similar Embedded designs typically do not need the flexibility and cost associated with sockets and so tend to connect SDRAM memory devices on the PCB SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc SDRAM Component Selection For the purposes of this document both have been considered since the effects of the DIMM socket are very minimal and can be ignored Therefore the issues to be examined relate only to the type and number of memory devices present Figure 5 shows atypical SDRAM module a similar arrangement would be used for embedded applications except perhaps for the serial presence detect SPD EEPROM SDCAS CKE WE MA 0 12 SDRAM Memory Bank 2 DQM 7 0 64 72 bits 4 18 Devices CS 7 0 SDRAM Memory Bank 1 64 72 bits 4 18 Devices D 0 63 SPD 0 2 Figure 5 Typical SDRAM Module Block Di
16. AN1722 D Rev 1 1 6 2003 SDRAM System Design Using the MPC 106 by Gary Milliorn RISC Applications Freescale Semiconductor Inc This document discusses the implementation of an SDRAM based memory system using the MPC106 The MPC106 PCI Bridge Memory Controller provides a bridge between the Peripheral Component Interconnect PCT bus and Freescale s MPC603e MPC740 MPC750 MPC745 MPC755 MPC7400 and MPC7410 PowerPC host processors To locate any published errata or updates for this document refer to the website at http www freescale com SPS PowerPC 1 1 Overview There are numerous possibilities available in designing systems although most will probably fall into the typical category shown in Figure 1 This document refers to components in Figure 1 when describing the issues involved in creating a PowerPC based system Note that many other configurations are also possible conductor Inc 2004 All rights reserved Ma e f 1 For More Information On This semiconductor Go to www freescale Freescale Semiconductor Inc System Analysis MPC106 Grackle rocessor Clock Figure 1 Typical SDRAM System Block Diagram Note High speed design systems operate at 100 MHz or less To achieve such a high speed it necessary to consider each source of possible problems and aggressively pursue solutions An accurate simulation of the board taking int
17. D PowerPC 6037 RISC Microprocessor User s Manual Freescale MPC603UM AD PowerPC 6047 RISC Microprocessor User s Manual Freescale MPC604UM AD MPC620 L2 Interface Application Note John Coddington High Speed Digital Design A Handbook of Black Magic Howard Johnson and Prentice Hall ISBN 0 133 95724 1 Foundations of Microstrip Circuit Design T C Edwards John Wiley NY 1981 Appendix B Organization and Loading Factors Table 27 summarizes organization and loading factors of several popular SDRAM DIMM modules The information was gathered from manufacturer data books and websites and is believed but not guaranteed to be accurate SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc e le pz oc loe oe oe loz loz je je je je je o loe oe joe 06 106 SA A SA 81 4 ZLXNY GWee nooz L 0vAZZSAH e i e e e le 6 oz se se se se 91 9 9 lez 60 60 60 08 log se 66 08 08 A A 91 z 9XNv ND0z L OPAP9SAH e le l je l le 6 po o loe o os le le fs joe joe joe je loe loe 99 so 06 06 SA A SA
18. Figure 17 Clock Trace Compensation SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Clocks 1 8 2 Clock Errors One final consideration with regard to the accuracy of the clock signals is that clock drivers are not perfect and have their own sources of errors cycle to cycle jitter and pin to pin skew The jitter specification must be tightly controlled or some cycles may be shortened causing data loss Therefore tight jitter control is needed Pin to pin skew means that some clock pins will change state at a slightly different time from the rest of the pins this also has the effect of possibly shortening an SDRAM access time Both of these errors must be allowed for when calculating the timing accuracy of the system 1 8 3 Clock Drivers Table 14 lists compatible clock drivers for SDRAM based systems Note that since the processor MPC106 memory buffers and SDRAM are all 3 3V only LVTTL clock generators may be used SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Timing Analysis Table 14 Compatible Clock Drivers Clock Counts Clock Skew Max System to System Jite FOEK System ROI Syston io PCI Freescale MPC972 200MHz 8 4 350 ps 500 ps 100 ps Yes Freescale MPC950 180 MHz 4 6 375 ps 500 ps 100
19. Freescale Semiconductor Inc Board Technology Finally compute the capacitance of the trace which is simply ATRACE CrRACE Z WIDE 1 5 2 Example PCB Measurements Using the formulas provided plug in the correct parameters for the PCB Table 7 shows the typical parameters for a high speed board Table 7 Selected PCB Electrical Parameters Parameter Description Value Units Notes TH Trace Height above ground plane 0 005 in 5 mil TW Trace Width 0 005 in 5 mil TT Trace Thickness 0 00137 in 1 oz Copper Relative Electric Permeability 4 5 mE FR4 Material Reduce the equations to get the propagation delay 0 5 ETHIN 2 75 1 75 L Ves ETHIN 3 24 EZ E e 12 000575 0 04 1 i Continuing 45 1 7 INO 0005 4 6 0 005 EEFF 3 03 Once the permittivity of the trace is known the propagation delay can be computed 12 8472x 10 3 03 Arrace 147 47ps in Thus for each inch of trace on the board allow for 147 ps of delay best case Remember that only the blank PCB material is being considered Once loaded the propagation delay will only increase Continuing on with the effective trace width TWepp 0 005 1 25 0 00137 E T n i Jl T 0 00137 TWepp 0 007 SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Time of Flight and characte
20. RAM modules to increase margins would improve write cycles but make matters worse for read cycles This is yet another reason registered transceivers are so desirable for high speed designs the pipelining allows the offset clock to compensate for the unbuffered control signals 1 10 2 Data Parity Along with the 64 bits of data there are 8 bits of data parity that may be used though it is not required Although many transceivers provide multiples of 9 bits to easily handle parity these bits cannot be used to SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc The Data Path implement the data parity path due to the alternate function the MPC106 assigns to them the ROM address lines If the data parity were not controlled separately attempts to read the ROM would cause bus contention between the ROM address lines and the data parity signals even if the ROM does not implement parity most do not the transceiver will drive something onto the bus either ones zeros or the last valid input signal received Consequently implementing a data parity path requires the use of a separate transceiver as shown in Figure 22 74ALVCH16501 Memory Parity Bus Processor MPC106 Bus Figure 22 SDRAM Parity Bus The MPC106 uses the dedicated PPEN signal to control reads of the memory parity data The arrangement of the buffer is otherwise identical to the d
21. SO CS2 13 46 13 46 pF CS1 41 41 pF CS3 34 34 pF DQ 0 7 9 20 9 20 pF As seen in Table 5 the number of capacitive loads can be very high almost 100 pF in certain cases Appendix B shows that the high loads are typically caused by the use of 4 bit SDRAM memories at a minimum of 16 18 per bank the capacitance of the address and control signals which must be wired to each SDRAM in parallel adds up quickly The embedded board designer can correct for this problem simply by using wider devices Sixteen bit wide memories are generally preferred as the capacitance drops fourfold Desktop board designers however are faced with the dilemma of choosing between designing for the worst possible case possibly increasing the cost or restricting operation to a particular set of SDRAM modules limiting flexibility Software can enforce restrictions on the quantity and type of SDRAM modules installed by examining the SPD EEPROM for details Too many loads and the system could deliberately refuse to run alternately sufficiently clever software could degrade the software programmable timing registers of the MPC106 to allow for slower operation The remainder of this document concentrates on the worst case values to show the magnitude of the effort involved since reducing the loads will only improve the margins and reliability 1 4 2 Module Timing Table 6 shows an extraction of the AC timing parameters fro
22. agram Generally SDRAM modules are 64 or 72 bits wide and comprised of 4 to 18 separate devices For greater memory expansion the module can include two separate banks of memory which doubles the size and the signal loads The designer will have to examine the total loads present to determine the total number of DIMMs that can be accommodated If every type of DIMM module were to be supported it might be necessary to limit the number of DIMM sockets used conversely if only certain types were allowed more sockets could be supported 1 4 1 Module Loading Appendix B Organization and Loading Factors shows a compilation of SDRAM module loading unit loads and total capacitance Table 5 summarizes the appendix into the minimum and maximum loading which might be expected Table 5 SDRAM Module Loading Typical Case Worst Case Signal Modules Loading Modules Loading Units Min Max Min Max 0 21 98 21 98 pF CKEO 21 69 21 69 pF CKE1 69 69 pF RAS CAS WE 21 98 21 98 pF CLKO 21 41 21 41 pF SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc SDRAM Component Selection Table 5 SDRAM Module Loading continued Typical Case Worst Case Signal Modules Loading Modules Loading Units Min Max Min Max CLK1 41 41 pF CLK2 CLK3 40 40 pF DQMB 0 7 8 25 8 25 pF C
23. al Medium 0 056 0 012 ns pF Herculean 0 043 0 011 ns pF Zeus 0 030 0 010 ns pF SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc SDRAM Component Selection With this information the normal mode output valid time at 83 MHz for a signal with a 150pF load could be calculated at 12 6ns The chart in Figure 4 shows the relationships 16 0 a 9 Zeus 8 0 i a ue ge a a LI Aas ap B Herculean Normal Output Valid ns 0 50 100 150 200 Capacitive Load pF Figure 4 MPC106 83 MHz Output Valid Derating 1 3 4 Errata Compensation The MPC106 does not meet the hold times required for PCI refer to the errata for more information While this is not ordinarily relevant to an SDRAM design the solution to the PCI hold time problem requires adding delay to the MPC106 clock Adding this delay to the MPC106 alone would rob the SDRAM section of margin so the delay has to be added to all system bus components that is CPU cache SDRAM and memory buffers As shown throughout this document adding deliberate delay time to signals is common to timing problems It is important to mention here that this timing correction described in the errata is applied independently of any timing adjustments performed Section 1 12 Example System details this global change as well as the individual changes to be made 1 4 SDR
24. at these two frequencies will be successful The remaining allowances at 83 MHz and beyond are very small and we have yet to allow for PCB traces so clearly there are some obstacles to overcome Table 16 provides more specific information regarding TOF Margin Table 16 Margins Bus Speed Measurement Conditions 75 MHz 2 LL ewe ae 6 traces heavy load 0 1 6 2 0 2 7 3 4 0 4 Le As expected once the PCB traces are included through the TOF factor most of the table entries are no longer viable Clearly most designs are not suitable using the basic defaults specified series termination worst case loading normal clock routing Ways to overcome these problems are discussed in the rest of this section 1 9 1 Hold Time Before examining solutions to the setup time problems insure that the SDRAM input hold times are met Table 17 compares the MPC106 and SDRAM components requirements SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Timing Analysis Table 17 System Hold Time Analysis Measurement MPC106 CLK to output invalid output hold Ignore the TOF at this point The requirement here is that the SDRAM inputs must not change for 1 0 ns after the clock edge If the clocks are in sync then a long TOF only lengthens the hold time the SDRAM input will see no change until TOF 1 0ns In fact in Table 17
25. ata bus path It might appear that this arrangement will not allow writes to the ROM space but the MPC106 buffers the ROM write data to eliminate any contention Note that while the MPC106 can use the parity lines to implement ECC with fast page or EDO DRAM this is not possible in an SDRAM configuration The decision as to whether or not you implement parity should consider if it is worth the bother and expense 1 10 3 Eliminating Memory Buffers A commonly asked question is whether these expensive registered transceivers may be deleted or whether less expensive flow through transceivers may be used The answer in both cases is a qualified yes but there are important caveats to be aware of Consider that the MPC106 and the PowerPC processor expect SDRAM data to be transferred on each beat of the clock during burst transfers and for SDRAM every cycle is a burst transfer Even if the first byte of the first transfer is all that is needed the MPC106 must complete the remaining three cycles If no buffers are present the SDRAM must be capable of driving the data bus load in the time interval given The time is constrained by the bus loading which might be light or severe and includes such loads as the processor other processors the MPC106 other SDRAMs and L2 cache modules Additional constraints are that the SDRAM may not be able to drive much of a capacitive load A typical SDRAM module may only drive 2mA into a 50pF load The rise time
26. chosen SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Table 12 Termination Characteristics ee 2 2 Typical Termination Signal Point to Point Power Recommended For Type Rise Time Wiring bea Dissipation Parallel 1 1Z C Yes 120 mW gt 83 MHz Desktop Systems Series 2 27409 No 98 mW lt 83 MHz Low power Systems AC 1 1Z C Yes 41 mW Clock Signals The best termination for a specific application depends heavily upon the physical environment PCB size and various other factors 1 8 Clocks A critically important issue in the design of an SDRAM system is the clocking To get the high speeds every important part of the SDRAM array is clocked at the maximum transfer rate which is as high as 100 MHz Clearly these speeds require careful attention to loading and routing in order to assure that each part of the memory system is in sync An important consideration to keep in mind when designing a clocking scheme is that the MPC 106 is a fully synchronous design and synchronizes all CPU bus activity which includes memory accesses by multiplying its PCI clock input with a PLL to match the CPU bus This means that the MPC106 s clock must be skew controlled with respect to the CPU bus clocks even though they are not usually the same frequency group Most of the clock controllers on the market do not follow this spec
27. e memory control data bus PCI This document only takes into account the memory control bits although the other capabilities should be kept in mind when designing the rest of the system Table 1 shows the three valid impedance settings for the memory control I O drivers SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc MPC106 Memory Controller Table 1 MPC106 Memory Control Output Impedance MPC106 Output Drive Alias Software Control 20 Q Medium Normal OD CR bit 4 0 ODCR bit 0 1 130 Herculean ODCR bit 4 1 ODCR bit 0 0 Zeus ODCR bit 4 1 ODCR bit 0 1 To minimize power dissipation ringing EMI etc use the highest impedance Since none of these impedances match the general range of PCB impedances available 40 80 ohms either termination will be required or the traces will have to be extremely short Since this register is controlled via software it must be initialized soon after reset occurs and before the memory system is initialized or used This is a fairly standard practice with MPC106 based systems since the PCI drive level often needs to be specified as well 1 3 2 DC Timing Parameters The DC parameters are shown in refer to the latest hardware specifications for updates to this information These numbers will be used primarily to determine whether or not series or parallel termination res
28. e a minimum of 0 3 ns margin An alternate way to approach this not covered here is to select a particular clock offset and compare the margins remaining This is particularly attractive if the target system must operate at numerous bus speeds as with programmable PC motherboards Table 19 Margins Faster Modules Clock Offset The delays necessary to implement the clock offsets shown above can be implemented on a PCB by creating delay line traces and is discussed in the following section Note that in the 100 MHz case a 3 0 ns delay is required While this is certainly possible on a PCB it constitutes 3046 of the 10 ns cycle time at that speed just for a corrective factor alone That is a good hint that a heavily loaded bus may not be able to function at very high speeds 1 9 4 PCB Delay Lines To implement the delay on the PCB re use the equations from Section 1 6 Time of Flight Be careful to use the capacitance of the clock inputs which are not necessarily the worst case loads for a particular SDRAM module Additionally the clock inputs are often driven by dedicated signals and so are not necessarily affected by the number of modules though this depends upon the clock architecture chosen Table 20 illustrates a range of loading for clock signals from 25 to 55pF We can use these values to determine amount of PCB trace we will need to implement the necessary clock offset SDRAM System Design Using the MPC106 For More Informa
29. e most common types of high speed boards Note Four layer boards are electrically close but typically have severe routing limitations that make high speed board layout difficult to impossible TT TH LAYER 1 SIGNAL LAYER 2 GROUND TH LAYER 3 SIGNAL Z 50Q TW 0 008 TH 0 005 TT 0 0137 1 oz copper LAYER 4 SIGNAL LAYER 5 POWER LAYER 6 SIGNAL Figure 6 Typical PCB Stackup Diagram SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Board Technology Equations required to calculate propagation delay are shown in subsequent sections Table 9 provides the typical propagation delay vs trace width for the board stack shown in Figure 6 and similar types Recalculation may be necessary for different board stacks The following characteristics of the traces will also be discussed e Electrical permittivity e Characteristic impedance e Capacitance per inch e Propagation delay per inch 1 5 1 Microstripline Calculations The typical PCB stackup shown in Figure 6 produces traces that are routed over a single power or ground trace producing a microstripline as opposed to striplines which are surrounded by planes above and below The following formulas can be used to determine the propagation delay through the board First calculate the electric permittivity of the stripline on the circuit board Permittivity
30. e of the amount of time it takes for an electrical signal to propagate through a PCB trace The effects of capacitive loading will be discussed To determine TOF the following information is required e PCB characteristic inductance in e PCB characteristic capacitance in e SDRAM module capacitive loading SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Time of Flight e Number of SDRAM modules in parallel e Longest routed trace The first three parameters have been covered in the preceding sections The SDRAM loading must be the worst case loading on any particular control signal it does not matter what the lesser loaded signals do since the MPC106 presents each with identical timing if the worst case Time Of Flight works the best case will also The trace length presents somewhat of a problem as the PCB may not have been routed yet so trace lengths of 4 and 6 inches is assumed and the results are presented for each Later if there is a problem reduce the trace length or loads and recalculate these equations To calculate the TOF a good model of the capacitance of the trace is needed To this factor incorporate the capacitance of the pins of the SDRAM modules C Additionally the modules are not lumped together at one location but are connected in series If the modules are assumed to be evenly spaced as is typical with a memory design we
31. ell with the smaller modules which have no loading on CLK3 or CLK4 at all This technique works regardless of whether or not DIMMs are used if not just connect the clocks as shown in Figure 15 SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Clocks Low Skew Clock Driver Figure 15 Shared Clock Architecture Note that this example shows series termination for the clock lines To handle the reflections for this type of termination we would attempt to equalize the Y branch at the destination keeping each split of the trace as short as possible but also equal to the other branch While some reflection is inevitable this will minimize the effects during the settling of the reflected waveform In a manner similar to this one we can also share the outputs of the clock driver as shown in Figure 16 By splitting clock trace near the clock driver and routing two equal length traces we can drive one SDRAM module with one clock pin This method requires a clock driver with a sufficiently low output impedance and special changes to the series termination Low Zp Low Skew Clock Driver Figure 16 Multiply Shared Clock Architecture SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Clocks The board impedance can be calculated using the formulas discussed in Sec
32. erratic operation or causing a loss of performance because compensating delay cycles must be added to the controller initialization It is strongly encouraged that boards be simulated with real data as much as possible Data should include the following e IBIS Spice models for the MPC106 and SDRAM memories e Spice models including trace lengths for the JEDEC memory modules e Estimated or actual trace lengths e Estimated or actual crosstalk due to adjacent traces 1 7 1 No Termination Itis possible to ignore termination only if the traces are very short this relationship is described and derived in the book High Speed Digital Design A Handbook of Black Magic refer to Appendix A Bibliography for more information The maximum trace length is constrained by the relationship Line Delay lt 2 so that for the MPC106 where t is 1 12 ns unterminated traces should have no more than 0 166 ns of delay compare this to the calculation of propagation delay discussed previously or approximately 1 inch This is extremely difficult to accomplish without using MCMs 3D chip stacks or other exotic packaging so most systems will need to use some sort of termination to compensate for the mismatch between the circuit board impedance 40 80 ohms and the output impedance of the MPC106 8 20 ohms Fortunately this deliberate mismatch can be easily compensated for with termination resistors and in addition will help clean up the signals
33. es 6 traces light load load _ traces heavy load 6 traces heavy load This has opened up a few cases at 75 and 83 MHz but the higher speeds are tantalizingly out of reach A glance at Table 15 yields no interesting ideas Tightening up the clock specifications would obtain given an absolutely perfect clock a margin gain of only 800 ps scarcely enough for a lightly loaded 90 MHz bus and nothing else Clearly the clock to output time Tcro of the MPC106 is the stumbling block here If we could do something about Tcro we might have something 1 9 3 Clock Offset The final approach to increasing margin addresses the problem of the clock to output time Tcro of the MPC106 Clearly we cannot change the chip to alter these fundamental timing values so instead we approach from a different direction Until now we have assumed that the MPC106 the CPU and the SDRAM were all exactly synchronized with the usual error allowances However by introducing a deliberate clock offset to the SDRAM all of its timing parameters will be offset with respect to the remainder of the system Figure 19 shows the standard MPC106 SDRAM timing waveform except that the SDRAM now receives a slightly delayed clock SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Timing Analysis Tor overlaps Tsy L OU yy E Tor _ a e Tsus
34. gnals MA 0 12 SDRAS SDCAS DQM 0 7 CS 0 7 WE CKE can be routed These signals should all have equal length traces as well but the manner SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Freescale Semiconductor Inc Physical Layout in which they are routed is affected by the choice of termination selected There is no need however to add deliberate delay to the traces as with the clock If series termination is used be sure that there are no stubs along the trace If the trace drives one load or pin this is easy however many applications will want to drive multiple pins on one or more DIMM modules with one signal This will be inevitable as there is only one CKE pin yet most DIMMs require two refer to Figure 23 Delay equalizes to second trace DIMM 1 DIMM 2 Equal length Stubs to MPC106 lt Figure 23 Series Termination Routing For series terminations the series resistors are placed close to the MPC106 pin then equal length traces extend out towards the DIMM array Equal length traces are also used when one trace is used to drive two loads as with the CKE signal example For such a case the trace is split at the destination into two equal length stubs It is not possible to eliminate reflections in these cases but it is possible to minimize them by using equal loads in the DIMM sockets wit
35. h equal stub traces so that the reflected waveform happens at identical times with identical energy Conversely if parallel termination is used route the signal point to point see Figure 24 this is much easier for layout The termination should be placed at the end of the trace with a minimum connection DIMM 1 DIMM 2 to MPC106 Figure 24 Parallel Termination Routing 1 11 3 Data Path Routing Regardless of the presence or absence of the data bus transceivers point to point wiring may be used If data bus registers are used reflections caused by the stubbed bus are quelled before the data is latched if registers are not used the bus is and must be operated at a sufficiently slower speed to attain proper operations SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Example System 1 12 Example System This section describes an example MPC106 based SDRAM system It is not a full schematic diagram detailed schematics are available at the Freescale PowerPC web site Refer to http www freescale com SPS PowerPC and look in the Technical Support section for the latest information on the Yellowknife project The following describes the connection between the MPC106 and the SDRAM memories It is very straightforward if the following things are kept in mind e modules are little endian PowerPC is big endia
36. have the following model of our SDRAM signals shown in Figure 7 Co _ y ee b d No Figure 7 Distributed Capacitance Using this model use the following equation to calculate the total capacitance seen per unit time No C Co egt length This equation says that the total capacitance is that of the trace Cj plus that of a quantity No of pin capacitive loads distributed along the PCB trace length In turn the delay of the trace can be computed by the usual electromagnetic wave equation but scaled into picoseconds inch units A 1x10 L5 Co ps in Table 9 shows representative C values for three different loading factors Heavy Moderate and Light best summarized as the total amount of capacitive loading presented For each of the two trace SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Time of Flight lengths considered the capacitance varies fourfold Table 9 Effective Capacitance for Various Design Factors Co per Trace Length Loading Number of C Per C Total Units Category DIMMs DIMM L m Heavy 2 98 pF 196 pF 51 27 34 93 pF Moderate 3 45 pF 135 pF 36 02 24 77 pF Light 1 60 pF 60 pF 17 27 12 27 pF Using these values and the board inductance calculation Lo it is possible to calculate the propagation delay Then obtain
37. ife or for any other application in which the failure of the Freescale Semiconductor product could create a situation where personal injury or death may occur Should Buyer purchase or use Freescale Semiconductor products for any such unintended or unauthorized application Buyer shall indemnify and hold Freescale Semiconductor and its officers employees subsidiaries affiliates and distributors harmless against all claims costs damages and expenses and reasonable attorney fees arising out of directly or indirectly any claim of personal injury or death associated with such unintended or unauthorized use even if such claim alleges that Freescale Semiconductor was negligent regarding the design or manufacture of the part e se 2 freescale semiconductor For More Information On This Product Go to www freescale com
38. ification in fact most typically allow the PCI clocks that is the MPC106 clock input to lag the CPU clocks by 1 4 ns MPC106 based systems must select clock chips which can generate a bus frequency 66 83 100 MHz synchronized within 250 ps of the PCI frequency 25 33 MHz Thus the basic clock architecture builds on the basic diagram shown in Figure 12 SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Clocks 50 100 MHz 2 2 5 3 V V V PowerPC L3 Cache MPC106 PCI CPU Memory Devices SDRAM CPU Clocks PCI Clocks 50 100 MHz 25 33 MHz Figure 12 Basic Clock Generation Each SDRAM memory component requires a clock with clock skew controlled to within the tolerances for setup and hold DIMM modules have from 4 to 18 SDRAM memories on them controlled by 1 to 4 clock pins The clock system for SDRAMs attached to the system board can be controlled by the clock system by balancing the loads and distributing them equally DIMM modules however introduce an additional complication because it is unknown what sort of module will be installed so a clock system that controls any possible module must be designed Further complicating matters is that different module manufacturers have different approaches as to how clock lines are connected on the module Table 13 shows all known clocking schemes SDRAM System Design Using the MPC1
39. ign of a series termination is fairly simple the value chosen for Z is such that it plus the output impedance 24 equals the board impedance Z As an example if the MPC106 is configured for 8 Q outputs and the PCB is designed with 50 Q traces then the series termination resistor should ideally be 42 Q Series terminated signals will typically dissipate much less power than a parallel terminated equivalent For such circuits power is dissipated only during the interval when the signal propagates from the source to the destination and back at which time the signal is critically dampened Once the net quiesces no additional power is drawn because both the source and the destination are at equivalent voltage levels Series terminations are therefore ideal for low power applications such as notebook computers and certain classes of embedded controllers To calculate the power dissipations assume that the net has been charged to 3 3V or ground 0V and that the MPC106 will drive the net to the opposite state not or Von which are minimums Calculate the currents needed and insure that they do not violate the specification of the MPC106 and that the power dissipation of the resistor is not exceeded Using the previous example the respective Io or Ioh values would be 78 mA and the 255 mW These numbers exceed both the MPC106 specification on current sink source and the maximum power dissipation in a typical resistor network or disc
40. istors would cause excessive current demand in the I O driver Table 2 MPC106 DC Electrical Characteristics Output high voltage lon 18 mA Output low voltage Output low current Vol 0 5 V MPC106 Output Drive 1 The loh and lo values are dependent upon the strength of the driver selected 2 Shaded items are derived and are not tested or guaranteed SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc MPC106 Memory Controller 1 3 3 AC Timing Parameters The AC timing information shown in Figure 3 and Table 3 may also be obtained from the MPC106 hardware specifications uut WLLL XE 90 7 N 10 E ty ti Figure 3 MPC106 Timing Parameters Table 3 MPC106 Memory Timing Parameters Timing Maximum Description Clock to output SDRAM These values are specified at a load of 50pF If the load is different than this reference point then the output timing values will have to be derated using the following formula t actual table 7 tractor The actual timing may be derived from the value in the timing tables of the MPC106 hardware specifications as shown in Table 3 along with the capacitive load of the design and the timing derating factor shown in Table 4 Table 4 MPC106 Timing Derating Factors Drive Strength Output Valid Factor Output Hold Factor Units Norm
41. m several manufacturer s data sheets To date all of the SDRAM modules and discrete components will work provided these timing values are allowed for Some manufacturers do much better on these such as requiring less setup time Although it will be easier to design for those parts the board will be restricted to only those parts This is easier to allow for in an embedded design as opposed to a desktop motherboard design where the manufacturer of the module may not be known SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Board Technology Table 6 SDRAM Timing Parameters Maximum Time ns Name Description 66 MHz 83 MHz 100 MHz 125 MHz tiss Input setup time SDRAM 3 0 3 0 3 0 3 0 tins Input hold time SDRAM 1 0 1 0 1 0 1 0 tohs Output hold SDRAM 1 0 1 0 1 0 1 0 ls Rise time SDRAM 1 0 1 0 1 0 1 0 tfs Fall time SDRAM 1 0 1 0 1 0 1 0 1 5 Board Technology The printed circuit board holding the components also contributes to the delay between the MPC106 and the SDRAM There are numerous factors which all vary widely so each will have to be considered abstractly Some of the factors to consider are e Trace width e Trace height over ground planes e Trace thickness copper thickness e Material used for board e Stackup organization Figure 6 shows a typical 6 layer stack up for a PCB perhaps one of th
42. n modules pack bank select lines at the top of the used address lines The first item means that the address line connections will look backwards If it was not for the fact that address ordering is significant in that they are present or absent depending on the size of the module this distinction could be ignored and the address lines could be hooked up in any order As it is though the address lines will not have to be transposed Further complicating matters is that big endian machines must renumber address lines when additional address bits are added as was done between the MPC106 V3 and MPC106 V4 refer to MPC106 Revision 4 0 Supplement and User s Manual Errata The second item is the packing of the bank selects the MPC106 takes care of this internally on the memory address lines MA 0 12 but this means it will appear that the bank select lines are treated as memory addresses which they are in a sense The MA bus is so flexible that assigning a name to the pins is rather difficult if you are confused refer to the MPC106 PCI Bridge Memory Controller User s Manual for diagrams and descriptions Table 24 shows the important connections between the MPC106 and a standard JEDEC SDRAM DIMM module Table 24 MPC106 SDRAM Interface Connections SDRAM 168 Pin JEDEC DQM 0 7 J15 H15 G16 E16 28 29 46 47 G14 G13 F14 E14 112 113 130 130 30 amp 45 1st socket Bank 0 m ens eem E
43. o account the trace lengths and trace separations is essential For designers who are considering lower speed memory systems it is possible to ignore some of these considerations 1 2 System Analysis Important factors to consider when designing an SDRAM based system follows e MPC106 AC timing e MPC106 output impedance settings e SDRAM setup hold time e Time of Flight TOF The MPC106 timing parameters are generally a function of the operating speed and are fixed however other attributes such as output impedance are configurable and timing relationships between the MPC106 and memories can be achieved by careful adjustments to the clock The SDRAM components are also generally constant in setup hold times across many vendors The contribution of capacitive loading from a typical DIMM socket is negligible so this document refers to SDRAM modules when referring to the memory Note that embedded applications using discrete SDRAM memories are equally viable The last factor is the Time of Flight TOF which refers to the amount of time it takes for a signal to propagate from one point to another Unlike the other factors TOF contains several subordinate factors that have to be considered and is also affected by the electrical characteristics of the MPC106 and the SDRAM Figure 2 shows an idealized signal connection between the MPC106 and an SDRAM component The term Zo for example is commonly used for both board impedance and ou
44. of such a driver if it works at all may require so many clock delays to compensate that the SDRAM would perform no better than EDO DRAM As for using less expensive buffers consider Table 21 which shows the propagation times for various 3 3V transceiver components SDRAM is not 5V tolerant so low voltage swing drivers are required SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Physical Layout Table 21 Representative Memory Transceivers hel SN74LVC16245A SN74ALVCH16501 SN74ALVCH16525 SN74ALVCH16543 SN74ALVCH16601 SN74ALVCH162601 If any non registered transceiver is inserted in the data path it adds to the TOF calculation when determining the margins for a design Recalling that the margins were very small at higher frequencies note that adding a flow through buffer can actually be worse than no buffer at all 1 10 4 Transceiver Selection Table 22 summarizes the choices that are best for various types of systems Note that these selections are not guaranteed Table 22 Memory Transceiver Selection Guide Recommended Buffer Capacitive Load Bus Speed Type Example High or Low 283 MHz Registered 16501 16601 162601 High lt 66 MHz Flow through 16245 162245 Low lt 66 MHz None 1 11 Physical Layout Previous sections have focused on the electrical conditions which must be met to insu
45. olled by the BUF_MODE configuration bit in the MCCR2 register The MPC106 enables either one to transfer data during read or write cycles Most transceivers also allow separate clocks for the read and write paths allowing finer tuning of the read and write propagation times though this is not usually necessary These registered transceivers handle the propagation delay by introducing a pipeline effect where the data is actually latched on the following cycle The MPC106 compensates for the extra clock cycle by overlapping new SDRAM cycles with the last cycle of the previous one so no penalty is incurred for the pipeline For these reasons registered transceivers are recommended for high speed designs and where capacitive loading may be high 1 10 1 Data Path Timing As noted in the previous sections the data path timing is affected by the usual board propagation factors plus the clock offset which may have been created to improve timing margins as described previously It should be obvious that delaying the clock to the SDRAM may improve timing for CPU to SDRAM write cycles but then the possibility arises that the return path may be damaged Examining the typical clock to output of an SDRAM module see Appendix B Organization and Loading Factors we see that clock to output of an SDRAM Tcos is remarkably similar to that of the MPC106 for each speed category This means that if no buffers are used then delaying the clocks of the SD
46. ps No Freescale MPC980 66 MHz 6 4 350 ps 350 ps 150 ps No ICWorks W42B972 200 MHz 8 4 350 ps 500 ps 100 ps Yes MicroLinear ML6500 80 MHz 4 4 300 ps 300 ps 150 ps Yes TI CDC586 CDC2586 100 MHz 6 6 500 ps 500 ps 200 ps No AMCC SC3506 80 MHz 10 10 500 ps 500 ps 0 No 1 9 Timing Analysis Figure 18 shows how the previously selected timing factors will be used together CLK e Tcrac 9 o Tone OUTPUT gt sus t Ru LLL Toros m Tons LLL KZ MARGIN Note The timing gap Margin is the total allowance for errors due to system noise clock jitter clock skew and a host of other problems The greater the margin the greater the reliability and robustness of the SDRAM memory system Figure 18 MPC106 Timing Analysis Since TOF is dependent upon the trace length which may not be well known in advance a useful first step in checking the timing is to compute the TOF and margin as a remainder of the other timing factors see Table 15 SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Timing Analysis Table 15 TOF and Margin Remaining Bus Speed Measurement MPC106 CLK to output valid 15 2 An examination of these numbers discloses a large allowance for margin and TOF at 66 and 75 MHz soa reasonable expectation is that the operation
47. re proper operation SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Physical Layout with scant attention paid to physical layout other than restrictions on the line length This section considers some of the mechanical properties that will affect the layout This section will not be very detailed because the possible number of layouts dwarfs the number of electrical choices 1 11 1 Clock Routing Careful attention must be paid while routing the clock signals In order to operate this system at 100 MHz apply the skew controls to the clock signals as discussed in Section 1 9 Timing Analysis To show the overall relationships needed assume a complete 100 MHz SDRAM based system with PCI slots and L2 cache and apply the MPC106 errata and the adjustments to the clocks This produces the overall adjustment shown in Table 23 Table 23 Clock Route Adjustments Clock Signal Reference PCI Slot Socket MPC106 Errata Clock Offset Total Point Allowance Allowance Adjustments Adjustments CPUCLK 42 5 1 0 6 0 Ins 75 MPC106CLK 42 5 38 0 1 4ns 10 5 L2CACHECLK 2 5 36 0 Tns 48 5 SDRAMCLK1 42 5 2 5 6 0 ins 16 ns 422 SDRAMCLK2 42 5 2 5 6 0 ins 16 3ns 422 BUFFERCLK i n 42 5 16 0 1ns 18 5 PCICARDS 1 n REF 4
48. rete component usually around 100 mW These are however peak currents and as noted previously current flows only during the propagation of the signal on the net Therefore the current and the average power dissipation are both scaled by the proportion of the bidirectional reflection of the signal to the clock period Imax 2 TOF 2 TOF Ilave Pave Assume 75 MHz system bus a TOF of 2 5 ns as discussed previously to get an average current draw of power dissipation of 30 mA and 98 mW acceptably within the limits Thus while the peak current needs are the same as a parallel circuit the average current needs are smaller SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Termination Because series terminated circuits are stable until the reflected wave has completed wiring signals in a point to point manner creates problems The preferred wiring for series terminated circuits is to connect only one load to the network If the signal must be shared connect the loads using equal length traces from a short common point near the destination This is discussed in Section 1 11 Physical Layout Series terminated circuits are the easiest to design and they are attractive in terms of power dissipation and cost Their drawbacks include an increased rise time two times that of parallel AC methods and the inability
49. ristic impedance Z 120 1 WIDE 7 0 007 0 007 1 393 0 667 1 1 444 3 03 se ae a i Zwipg 63 830 The impedance is important to know later when signal termination is discussed When PCBs are purchased the impedance above will be the ideal goal for the manufacturers to attain Finally compute the capacitance of the trace which is simply _ 147 47 CrRACE 63 83 Crrace 231 pF in These calculations are used to calculate the timing losses Remember that all of these are ideal values from which others may be derived In particular there are no loads attached and SDRAM loads can be quite severe 1 5 3 Reference Values Table 8 allows ideal propagation delay and other important calculations to be read from the trace width and height assuming FR4 material and 1 oz copper traces Table 8 Ideal PCB Electrical Parameters Impedance Capacitance Propagation Delay 0 005 0 005 63 83 9 97 nH in 2 27 pF in 147 40 ps in 0 006 0 008 56 50 8 50 nH in 2 67 pF in 150 43 ps in 0 010 0 010 65 90 Q 9 87 nH in 3 02 pF in 152 17 ps in The typical boards are all reasonably close in propagation delay impedance and capacitance values If the board routing choices are not known in advance use any of the above and the final calculations will not change by much 0 006 0 010 50 40 Q 11 39 nH in 1 95 pF in 152 17 ps in 1 6 Time of Flight This section will determine the Time of Flight TOF which is a measur
50. s the PowerPC processor maintains separate tenures exclusive use of the address and data bus so too does the SDRAM controller in the MPC106 separate the SDRAM control and data bus Indeed the MPC106 memory controller does not intercept the SDRAM data in any way it merely enables the buses and expects the processor to read or write the data at the appropriate time Our job on the data bus then is to design a high speed bidirectional data bus One saving grace of the SDRAM memory array is that each bank of memory typically contributes one or two capacitive loads on each data bus signal unlike the address and control signals which could add as many as 18 loads per bank A confounding factor however is the presence of unknown parasites on the data bus such as cache modules or other processors in a multiprocessing system So the external environment beyond the SDRAM array now becomes an important factor To achieve high speed SDRAM designs we will typically need registered transceivers between the SDRAM data bus and the processor data bus An example of the data path is shown in Figure 21 SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc The Data Path B to CPU to SDRAM MRCLK MWCLK Figure 21 SDRAM Data Path Element The transceiver has two separate output enables which are controlled by the MPC106 operating in the RE and WE strobe modes contr
51. scaleSemiconductor hibbertgroup com AN1722 D Information in this document is provided solely to enable system and software implementers to use Freescale Semiconductor products There are no express or implied copyright licenses granted hereunder to design or fabricate any integrated circuits or integrated circuits based on the information in this document Freescale Semiconductor reserves the right to make changes without further notice to any products herein Freescale Semiconductor makes no warranty representation or guarantee regarding the suitability of its products for any particular purpose nor does Freescale Semiconductor assume any liability arising out of the application or use of any product or circuit and specifically disclaims any and all liability including without limitation consequential or incidental damages Typical parameters which may be provided in Freescale Semiconductor data sheets and or specifications can and do vary in different applications and actual performance may vary over time All operating parameters including Typicals must be validated for each customer application by customer s technical experts Freescale Semiconductor does not convey any license under its patent rights nor the rights of others Freescale Semiconductor products are not designed intended or authorized for use as components in systems intended for surgical implant into the body or other applications intended to support or sustain l
52. ster clock These chips are often referred to as robo clocks because they do no synthesis but only replicate the input to multiple outputs This architecture can benefit boards with routing density problems typically densest is the CPU MPC106 area by requiring only one clock line to be routed over to the SDRAM array This also frees up many other clock outputs to be used elsewhere The disadvantage of course is the additional cost of the robo clock SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Clocks Low Skew Clock Driver Robo Clock Driver Delay Compensation Feedback Figure 14 Distributed Clock Architecture If the above examples are not possible usually due to an insufficient number of clock outputs and or cost problems then compromises in the clock generation and attempts to compensate for them must be explored In each of the remaining cases the capacitive loading increases and so the propagation delay does also This means that the timing must be carefully analyzed in light of the new side effects One possibility is to share clock lines for the SDRAM modules Table 13 shows several combinations One good possibility is to combine CLK1 with CLK3 and CLK2 with CLK4 This works well with the 72 bit wide modules distributing the clock lines with 5 loads so that no clock line must drive more than one such load It also works w
53. the TOF time is the timing margin 1 9 2 Insufficient Margin If because of heavy loading by the SDRAM parts or PCB layout restrictions there are no margins or perhaps the margins do not meet the basic TOF requirements some additional work will have to be done If on the other hand the margins are acceptable for the system being designed go directly to Section 1 8 3 Clock Drivers If more margin is needed first try changing the definition of the problem so the problem solves itself In this case restrict the type and or number of SDRAM modules allowed to be installed embedded systems designers get to do this for free just by changing the bill of materials While this might seem like cheating Appendix B Organization and Loading Factors shows that the most heavily loaded modules dual bank DIMMs and those which use 4 bit wide memories restricting the free use of those particular types can reduce capacitive loading by one half to one fourth Take for example a Samsung KMM366S824AT 64 Mbyte 1 bank x16 SDRAM module where the input setup time is reduced to 2 0 and 2 5 ns at 83 MHz and 100 MHz respectively A revised margin table is shown in Table 18 SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Timing Analysis Table 18 Margins with Faster Modules Bus Speed Measurement Conditions TOF TOF Margin 6 trac
54. tion 1 5 Board Technology As an example the Freescale MPC972 clock generator meets the general requirements of a PowerPC clock source and additionally has a 79 output impedance The relationship is determined by the following equation So plugging in what we know 79 20 ZQ 500 500 Or simplifying ZyQ 2 25Q 7Q 360 1 8 1 Clock Offset At this point the clocks are wired to the SDRAM but there is one other factor to consider the capacitive loading on the clock signals and its effect on the clock Worst case loading will be around 40pF per clock signal refer to Section 1 4 SDRAM Component Selection for more information When we previously calculated TOF values we used the worst case loading factors for signals such as SDCAS Clocks tend to be loaded somewhat less since there are more of them so recalculating the TOF factors for the clock TOF cx gives a faster propagation delay Thus the clock arrives significantly before the signal robbing the SDRAM of the hold time that is normally present These differences in the Time of Flight will need correction so the routing of the clock signals will need adjusting relative to the more heavily loaded control and data signals refer to Figure 17 In fact a deliberate offset to the clock signals usually via additional trace length may be needed to achieve high speed operation Uncompensated Clocks MPC106 Compensated Clocks MPC106
55. tion On This Product Go to www freescale com Freescale Semiconductor Inc Timing Analysis Table 20 Clock Offset Trace Length Calculation mum pee serene es m ae Ten These trace lengths are quite long but are easily implemented on a PCB if sufficient routing area is available Figure 20 shows a delay line which can be implemented on a typical FR 4 PCB Take the 8 9 inch delay line for the moderately loaded 100 MHz system from Table 20 for example Assuming 50 PCB traces lay out the delay line as shown in Figure 20 puru aoe 1 If PCB is 0 010 TW 2 x TH 0 020 2 Add extra trace separation to reduce crosstalk TW 0 010 3 Center to center dimension is then 2 x TH 0 010 0 050 0 050 0 600 4 Pick a width to fit the available space 0 600 5 Number of crenelations Length 0 600 x 2 0 0500 x 2 Figure 20 Delay Line The above example would give the following delay line parameters Copies of delay segment 7 Total Area 0 7 L x 0 6 H 0 42 in When designing the delay line keep in mind that the accuracy inaccuracy of PCB fabrication houses and natural temperature variances of FR 4 PCB materials can cause the accuracy of the delay line shown in Figure 20 to vary by 10 1 9 5 Determining How Much Margin So far we have designed with a minimum target of 300 ps This figure was arbitrarily chosen to provide a 10 margin at 100 MHz
56. tions and so there are potentially numerous solutions any that meet the requirements would work as well For example Table 11 shows some selected resistor pairs and the resulting conditions Table 11 Parallel Termination Characteristics Case Number Vmid Case 3 provides more balance in the dissipation of power while case 2 maintains a steady state level above SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc Termination the active high threshold Case 1 is a compromise between the other two Table 11 also shows the power dissipation required for each case This power is always dissipated since there will be a current between any voltage on the net and one or both of the resistors The value received in exchange for this power dissipation is that the rise times of signals on parallel terminated nets is fast there is little to no reflection to affect the waveform Additionally since the waveform has full amplitude all along the net loads may be placed anywhere along the net with no distortion of the signal making it ideal for routing memory busses which tend to be very parallel and perfect for the sort of point to point daisy chain wiring that allows 1 7 3 Series Termination Series terminated systems place a series resistor in series with a signal near the driver pin as shown in Figure 10 Figure 10 Series Termination The des
57. tput impedance but it is used in this document exclusively for trace impedance SDRAM System Design Using the MPC106 For More Information On This Product Go to www freescale com Freescale Semiconductor Inc MPC106 Memory Controller MPC106 SDRAM TOF p gt Zp MPC106 Output Impedance Co Trace Capacitance Rs Series Termination Lo Trace Inductance Ry Parallel Termination Thevenin Trace Length C SDRAM Input Capacitance TOF Time of Flight Zo Trace Impedance Figure 2 SDRAM Interconnect Factors With the exception of the bidirectional data bus which will be discussed later the SDRAM devices are controlled exclusively with MPC106 outputs driving SDRAM inputs Thus the majority of the design can be completed using the model shown in Figure 2 1 3 MPC106 Memory Controller The MPC106 implements the entirety of the SDRAM controller with the exception of connecting the SDRAM data bus to the processor data bus discussed in subsequent sections It is important to have the latest revision of the MPC106 hardware specifications and the latest errata Only revision 4 0 of the MPC106 supports the timings needed for high speed designs as well as the programmable driver options For more information refer to Appendix A Bibliography 1 3 1 Output Impedance To support a wide variety of options the MPC106 has an internal register PICR1 which can be used to set the output impedance of various parts of the interfac
58. v9 1 06085995 E 2 uL E je e t e 6 joe jo jez lo je 0 0 oz o o o ce o se se se 1 1 1 v OLX Y9XWv lVverS99SINWM e i e je e l e ge e t le 6 pe o lss lsz 60 se or or jor jor loe loe 99 ss 06 06 A A AJA 91 exe YIXNY GZS NLGO0PS99SIAIO E z uL fe E pg le e 6 se se jse se lse 60 60 sec 0 or or lop loe 06 99 ss 06 06 A A A A 91 8XWc YIXNY GWee ZLIEOSIISNWNWA e k k je l e lb e t le 6 loz o Jor o jr lsz se se o 0 gr ier o oo o 09 o9 09 L L AJGA 6 ZLXNZ GIN91 NLBEOZSpZSWW e k 4 e M e pg e le 6 sc e sc sc sc lsz 6 sc o 0 Qr lop 99 99 fse se ss sS 1 L ALA 8 9IXNL Y9XWZ 8 9 NLGPOZS9SSWIN E z uL t e 6 joe jo jse lo je loz joz oz o 0 or 99 99 0 sc e sS L L AIA 8 exe Y XNZ GOL 860025995 e k k je l e jpg e le 6 loz jo lo je loz loz joz jo jo jo jr se se jo jse ise L L 1 8D NIGPOISOOSWWY v OLX POXINI SIN9 NLGPOS99SWN SL joe loe joe joc 2 62 lsz joe joe loe joe 86 se jos jos 66 lse A 81 8xWz vez L 0S 18 LA 01 jo joe joe fer ler ler se se pe je ps ps pos os os oS 1 1 AIGA 6 8XWc GING veZz 1 0S161N fsz lsz lsz 02 vi vi ge se 92 oz 88 68 68 88 L L A A 91 8xWz Y9XNv GgWee vvov L 0S19LLIN 01 Jo s 0 jc 8 8 8 0 60 sz 90 9 9 9 Sv r 19 LLL LALA 8 8XWc v9XWZ 8 9 vvoz L 0S181AN 8

Download Pdf Manuals

image

Related Search

Related Contents

スーパーロータリーST02    Gril turc traditionnel  Easy200/300 D/E/F/I/NL/ES 6.10  Manual  ROTEX Sanicube INOX  StarTech.com 15 ft Standard Computer Power Cord - NEMA5-15P to C13  Komplettgerät C7 613 - Service  風呂フタ  MANUAL DE USUARIO DEL PROGRAMADOR  

Copyright © All rights reserved.
Failed to retrieve file