ECE 485/585 Microprocessor System Design

Transcription

1 Microprocessor System Design Lecture 5: Zeshan Chishti DRAM Basics DRAM Evolution SDRAM-based Memory Systems Electrical and Computer Engineering Dept. Maseeh College of Engineering and Computer Science Sources: Lecture based on materials provided by Mark F. Jacob s DRAM Systems article Memory component datasheets

2 Outline Taxonomy of Memories Memory Hierarchy SRAM Basic Cell, Devices, Timing DRAM Basic Cell, Timing Memory Organization Multiple banks, interleaving DRAM Evolution DDR3 SDRAM DRAM modules Error Correction Memory Controllers

3 Dynamic RAM (DRAM)

4 DRAM Technology bit line DRAM Cell 1 transistor word line Bit state (1 or 0) stored as charge on a tiny capacitor Read is destructive must restore value Charge leaks out over time refresh Write Drive bit line Select desired word ( row ) Read Pre-charge bit line Select desired word ( row ) Sense charge Write value back (restore) Refresh! Periodically read each cell (forcing write-back)

5 Volatile Memory Comparison SRAM Cell DRAM Cell word line word line bit line Larger cell lower density, higher cost/bit Non-destructive Read No refresh required Simple read faster access Standard IC process natural for integration with logic Non-multiplexed address lines Density low enough to keep the number of address lines reasonable bit line Smaller cell higher density, lower cost/bit Destructive Read bit line Needs periodic refresh Complex read longer access time Special IC process difficult to integrate with logic circuits Multiplexed address lines Density >> SRAM density so would require lots of address lines

6 DRAM Device Pin Outs Cost Rules! Fewer pins, smaller package, less $ So Multiplex Data (In/Out) /WE asserted (low) for write /OE asserted (low) to enable output buffers Address (Row/Column) 256K = 18 address bits 9 row address bits 9 column address bits /RAS (Row Address Strobe) asserted after row placed on address pins Address /CAS (Column Address Strobe) asserted after column placed on address pins / 8 / 9 2Mb (256K x 8) DRAM Data (DQ) /RAS /CAS /WE /OE

7 DRAM Organization We want to keep row/column organization (square is good) but get devices of more than x1 bit wide (from Bruce Jacob)

8 Data Data In/Out In/Out Data In/Out Data In/Out Data In/Out Data In/Out Data In/Out Address Buffer Row Decoder Bit Line Bit Line Bit Line Bit Line Bit Line Bit Line Bit Line Bit Line Internal DRAM Organization 2Mb as 256K x 8 Square keeps the wires short: Power and speed advantages Less RC, faster pre-charge and discharge faster access time! A0 A8 9 Select the addressed bit 9 Memory Array Memory Memory (512 Array Array Memory Memory (512 x x 512) 512) (512 Array Array Memory Memory x 512) (512 (512 Array Array x x 512) 512) (512 (512 x x 512) 512) Word Word Line Line Word Word Line Line Word Word Line Line Word Line Memory Array (512 x 512) Word Line Sense Amps & I/O Column Decoder Do this 8 times Read 512 bits at a time Data In/Out D

9 DRAM Timing

10 DRAM Timing Read cycle - RAS + CAS (RAS asserted) Entire row is latched in a data register (CAS asserted) Data in register is multiplexed to output (RAS de-asserted) Data in data register is rewritten to row in array (CAS de-asserted) Output is released Write cycle - RAS + WE + CAS (RAS asserted) Entire row is latched in data register (WE asserted) Data to be written is stable (CAS asserted) Write Data to register (WE de-asserted) Data to be written is no longer stable (RAS de-asserted) Data in data register is rewritten to row in array (CAS de-asserted) Operation complete Refresh cycle - RAS Only (RAS asserted) Entire row is latched in data register (RAS de-asserted) The data in the register is rewritten to row in array

11 DRAM Read Timing Every DRAM access begins at: The assertion of the /RAS Two ways to read: early or late v. /CAS DRAM Read Cycle Time /RAS /CAS /WE /OE A 256K x 8 9 DRAM 8 D /RAS /CAS A /WE /OE Row Address Col Address Junk Row Address Col Address Junk D High Z Read Access Time Junk Data Out High Z Output Enable Delay Data Out Early Read Cycle: /OE asserted before /CAS Late Read Cycle: /OE asserted after /CAS

12 DRAM Write Timing /RAS DRAM WR Cycle Time /CAS A Row Address Col Address Junk Row Address Col Address Junk /OE /WE D Junk Data In Junk Data In Junk WR Access Time Early Wr Cycle: /WE asserted before /CAS WR Access Time Late Wr Cycle: /WE asserted after /CAS

13 Key DRAM Timing Parameters t RCD : RAS to CAS Delay Minimum time between RAS asserted and CAS asserted t CAC : Column Access Time Delay from falling /CAS to valid data out t RAC : Random Access Delay Determines Latency Minimum time from /RAS falling to valid data output Quoted as the speed of a DRAM t RAC = t RCD + t CAC t RAS : Row Address Strobe Minimum time /RAS must be maintained t RP : Row Pre-Charge Delay Minimum time to pre-charge before /RAS can be asserted again t RC : Row Cycle Time Determines Bandwidth Minimum time between successive row accesses t RC = t RAS + t RP

14 DRAM Evolution

15 DRAM Evolution A Technology Timeline FPM Fast Page Mode (< 1995) Other DRAM technologies EDO Extended Data Out ( ) Rambus DRAM (RDRAM) Concurrent Rambus DRAM BEDO Burst Extended Data Out Direct Rambus DRAM (DRDRAM) SDRAM Synchronous DRAM (>1995) SDRAM Numerous Specialty DRAMs not shown Virtual Channel Memory (VCDRAM) DDR SDRAM (double data rate) Enhanced SDRAM (ESDRAM) DDR2 MoSys 1T-SRAM DDR3 Reduced Latency DRAM (RLDRAM) Fast Cycle DRAM (FCRAM) DDR4 By 2002 most PCs were shipping with SDRAM and DDR SDRAM By 2010 PCs were shipping with DDR3 in volume and still are (from Bruce Jacob & David Wang)

16 Conventional DRAM Read Timing (from Bruce Jacob & David Wang)

17 Fast Page Mode DRAM Read Timing Innovation Hold entire row (page) in sense amplifiers Benefit CPU can access each column in row without providing row address (and pre-charging) each time (from Bruce Jacob & David Wang)

18 Extended Data Out DRAM Read Timing Innovation Add latch between sense amplifiers and output pins Benefit Can begin pre-charging sooner (data from prior access remains valid) (from Bruce Jacob & David Wang)

19 Burst EDO DRAM Read Timing Innovation DRAM provides column data sequentially Benefit No need to transfer column address on each read (from Bruce Jacob & David Wang)

20 Synchronous DRAM Read Timing Innovation Pipeline access, command interface (vs. individual signals) Benefit Simplified timing, command interface, (from Bruce Jacob & David Wang)

21 SDRAM (Synchronous DRAM) Adopted for Pentium use Synchronous (clocked) interface RAS, CAS, WE, CS signals combined to make command Burst read/write Initial latency, then data every clock cycle Internal interleaved banks allow multiple rows (pages) to be open for read/write Ideal for cache line fill when bus width < cache line size

22 SDRAM Details Multiple banks of cell arrays are used to reduce access time: Ex: Each bank is 4K rows by 512 columns by 16 bits Read and Write operations are split into row access followed by column access These operations are controlled by sending commands Commands are sent using the RAS, CAS, CS, & WE pins. Address pins are time multiplexed During RAS operation, address lines select the bank and row During CAS operation, address lines select the column. ACTIVATE command opens a bank/row for operation Transfers contents of the entire row to a buffer Subsequent READ or WRITE commands access the contents of the row buffer For burst reads and writes during READ or WRITE the starting address of the block is supplied. Burst length is programmable as 1, 2, 4, 8 or a full page (entire row) with a burst terminate option Special commands are used for initialization (burst options etc.) A burst operation takes 4 (address/read row) + n cycles (for n words)

23 Functional Block Diagram 8 Meg x 16 SDRAM 2 12 = 4096 rows (pages) 2 2 = 4 banks 2 9 = 512 columns

24 Key SDRAM Timing Parameters

25 Key SDRAM Timing Parameters t RCD : Determines Latency Minimum time between an ACTIVATE command and Read command Analogous to DRAM parameter t RCD : Row Command Delay (RAS/CAS Delay) CL: CAS Latency Determines Latency Time between READ command and first data valid Analogous to DRAM parameter t CAC : Column Access Time t RAS Time between ACTIVATE command and end of restoration of data in DRAM array Analogous to DRAM parameter t RAS : Row Address Strobe t RP Time to pre-charge DRAM array in preparation for another row access Analogous to DRAM parameter t RP : Row Precharge Delay t RC Determines Bandwidth Time between successive row access to different rows t RC = t RAS + t RP Analogous to DRAM parameter t RC : Row Cycle Time

26 CL CAS Latency SDR DRAM examples (DDR can have CAS latency of 2.5)

27 DDR (Double Data Rate) SDRAM Innovation Transfer data on rising and falling edges of clock Same internal SDRAM core but 2n-prefetch Benefit 2x the bandwidth, same control & signals as SDRAM Significant Differences: Source synchronous (DQS) Burst length of 2,4,8 only CL = 2, 2.5, 3

28 DDR (Double Data Rate) SDRAM (cont d) 2n prefetch Use same DRAM core (cell array) Fetch twice as many bits Same latency for first data transfer Source synchronous Data transfer is twice clock rate Data strobe sent alongside data Read: supplied by DRAM Data aligned with strobe edge Write: supplied by controller Data centered on strobe edge [from Elpida]

29 DDR SDRAM Access Examples Reads from same open page/bank Micron DDR Datasheet [From: Samsung]

30 Banks Incorporated Into SDRAM Memory address Row Bank Column Why row/bank/column, not bank/row/column? Consider spatial locality Imagine accessing a series of sequential memory addresses After exhausting a column, references to another bank Consider if row/bank reversed Bank would rarely be used, lose benefit of interleaving

31 DDR SDRAM Access Examples Reads from different banks, open row Row Memory address Bank Column [From: Samsung]

32 DDR SDRAM Access Examples Reads from different row Memory address Bank Row Column From: Samsung