CSEE 3827: Fundamentals of Computer Systems. Storage

Transcription

1 CSEE 387: Fundamentals of Computer Systems Storage

2 The big picture General purpose processor (e.g., Power PC, Pentium, MIPS) Internet router (intrusion detection, pacet routing, etc.) WIreless transceiver (e.g., wifi, iphone) addr data_out C L combinational logic sequential logic, finite state machines data_in r_en w_en Memory storage elements (net) D Q logic gates, boolean algebra flip-flops, latches

3 Registers A flip-flop can store 1 bit. A register is a set of n flip flops that stores n bits. 3

4 Register w. load control input (v1) Encapsulate logic inside register Send cloc signal to flip-flops only when Load=1 This technique is called cloc gating 4

5 Register w. load control input (v) Encapsulate logic inside flip-flop EN signal selects between current value of register (Q) or new value (D) Preferable to v1 as it leaves cloc signal unadultered. 5

6 Shift register A register capable of shifting bits laterally in one or both directions 6

7 Shift register w. parallel load Three modes: Serial input Parallel input No input 7

8 Shift register w. parallel load Serial input operation (enabled by Shift) 8

9 Shift register w. parallel load Parallel load operation (enabled by Load AND Shift) 9

10 Shift register w. parallel load Do nothing operation (enabled by Load AND Shift) 10

11 Ripple Counter D X C0 D Y C1 D Z C Each flip-flop feeds Q into D: complements its value X s Q feeds into Y s cloc, Y s Q feeds into Z s cloc - why? 11

12 Ripple Counter D X C0 D Y C1 Y s cloc D Z C Note: each FF is negative-edge triggered 1

13 Ripple Counter D X C0 D Y C1 D C Z C C C

14 Ripple Counter with Positive Edge Triggering D X C0 D Y C1 D Z C C C C

15 Serial Counter (counting upward) D X C0 C C C Y(t+1) differs from Y(t) only when X(t)=1 D Y C1 Z(t+1) differs from Z(t) when both X(t)=1 and Y(t)=1 so Y changes when X (evaluates to TRUE), Z changes when XY (evaluates to TRUE) X = X, Y = Y X, Z = Z XY D Z C for the jth bit (in a j+1 bit counter), Bj = Bj Bj-1Bj-...B1B0 15

16 Register MUXing Reg 00 In En Reg 01 In En In1 In Some -input, 1-output combinational circuit (e.g., +, -) Reg 10 In En Reg 11 In En Suppose have 4 -bit registers with Enable (for writing) combinational logic to perform some operation OP Goal: Allow system to select registers registers selected for inputs (A & B) 1 register for output (C) At end of cycle: C = A OP B Note: A&B can be same register, C can also be same as A or B 16

17 Register MUXing eample e.g., Reg01 = Reg01 OP Reg10 Reg10 = Reg10 OP Reg10 (if OP were +, this would do Reg10 *= ) OP selecting: be able to choose from different OPs, e.g. Reg01 = Reg01 + Reg10 Reg10 = Reg01 * Reg00 17

18 Register MUXing Reg 00 In En C A B SEL SEL -to-4 DEC Reg 01 In En Reg 10 In En 4-to-1 MUX SEL 4-to-1 MUX In1 In Some -input, 1-output combinational circuit (e.g., +, -) Reg 11 In En 18

19 Register MUXing Reg 00 In En C A B SEL SEL -to-4 DEC Reg 01 In En Reg 10 In En 4-to-1 MUX 4-to-1 MUX In Some -input, 1-output combinational circuit (e.g., +, -) Decoder selects In1 which register is enabled (will be written to) SEL Reg 11 In En 19

20 Register MUXing Reg 00 In En C A B SEL SEL -to-4 DEC Reg 01 In En Reg 10 In En Reg 11 In En 4-to-1 MUX SEL 4-to-1 MUX In1 In Some -input, 1-output combinational circuit (e.g., +, -) Top MUX selects using A, determines which register feeds into In1. Bottom MUX does same using B for In. 0

21 Eample: Reg01 = Reg10 OP Reg01 Reg 00 In En SEL SEL -to-4 DEC Reg 01 In En Reg 10 In En 4-to-1 MUX SEL 4-to-1 MUX In1 In Some -input, 1-output combinational circuit (e.g., +, -) Reg 11 In En 1

22 Register MUXing and OP MUXing SEL -to-4 DEC Reg 00 In En Reg 01 In En Reg 10 In En SEL 4-to-1 MUX SEL 4-to-1 MUX C A B Op1 Op Op3 Opsel 4-to-1 MUX Op4 Reg 11 In En

23 Serial Adder Circuit C Clear SI B Shift Reg SO X Y Z Full Adder S C Serial Input RESET C Clear SI A Shift Reg SO D Carry C R CLOCK New Value pushed into A register Result pushed into B register If Shift Reg holds N bits, SUM starts calculating during N+1st cycle Done after N cycles

24 Serial Adder Circuit Eample: = 1100 RESET Serial Input CLOCK C Clear SI C Clear SI B Shift Reg A Shift Reg SO SO X Y Z D R Full Adder C Carry To time 7, just rolling values into registers by adding with 0 s (B output) Adding starts time 8 with least significant bits S C Time In A B S C X X X X XX X XXX X XXXX

25 Memory interface Stores data in word units A word is several bytes (16-, 3-, or 64-bit words are typical) write operations store data to memory read operations retrieve data from memory DataIn n Address Read Enable MEMORY words n bits per word An n-bit value can be read from or written to each -bit address n Data 5

26 Conceptual view of memory 6

27 Memory array architecture (1) Memory is a D array of bits. Each bit in a cell. :4 Decoder cell bitline bitline 1 bitline 0 11 wordline 3 Address 10 wordline 01 wordline 1 00 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 7

28 Memory array architecture () Address is decoded into set of wordlines. Wordlines select row to be read/written. Only one wordline=1 at a time. :4 Decoder bitline bitline 1 bitline 0 11 wordline 3 Address 10 wordline 01 wordline 1 00 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 8

29 Memory array architecture (3) When writing, contents of word written to bitlines. :4 Decoder bitline bitline 1 bitline 0 11 wordline 3 Address 10 wordline 01 wordline 1 00 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 9

30 Memory cell abstraction Cell is base element of memory that stores a single bit wordline bit bitline wordline bit bitline 0 Z Implemented with a tristate buffer. Value Z does not drive output wire to either a 0 or 1. Implementation of cell varies with type of memory. Copyright 007 Elsevier 30

31 Types of memory Random access memory (RAM) Volatile (no storage when power off) Read-only memory (ROM) Non-volatile (retains data when powered off) Fast reads and writes Historically called RAM because equal time to read/write all addresses (in contrast to serial-access devices such as a hard dis or tape). Somewhat misleading as ROM also can have uniform access time. Fast reads, writing is impossible or slow (again, misleading name) Historically called ROMs because written by permanently blowing fuses (so rewriting was impossible). Modern ROMs, such as flash memory in ipod are rewritable, just slowly. Dynamic RAM (DRAM) Cell stores data w. capacitors wordline bit Flip-flop bitline Static RAM (SRAM) wordline Cell stores data w. cross-coupled inverters bitline Register bitline ROM Mas-programmed (at chip fab) Hard Dis PROM Fuse- or antifuseprogrammed FLASH Electrically erasable floating gate with multiple erasure and programming modes Copyright 007 Elsevier 31

32 Volatile storage (RAM) comparisons Flip-flop SRAM DRAM Transistors / bit ~0 6 1 Density Low Medium High Access time Fast Medium Slow Destructive read? No No Yes (refresh required) Power consumption High Medium Low 3

33 Storage hierarchy Non-volatile CPU Registers Cache (SRAM) Main memory (DRAM) storage (Hard dis or solid-state drive) Fast access to small amount of data Slow access to large amount of data 33

34 Bottom-up eamination of SRAM circuits :4 Decoder bitline bitline 1 bitline 0 11 wordline 3 Address wordline wordline 1 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 34

35 Bottom-up eamination of SRAM circuits () Bits in cells (modeled by an SR latch) :4 Decoder bitline bitline 1 bitline 0 Address wordline 3 wordline wordline 1 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 35

36 Bottom-up eamination of SRAM circuits (3) Cells wired into bitslices. :4 Decoder bitline bitline 1 bitline 0 Address wordline 3 wordline wordline 1 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 36

37 Bottom-up eamination of SRAM circuits (4) Bloc diagram of a bitslice :4 Decoder bitline bitline 1 bitline 0 Address wordline 3 wordline wordline 1 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 37

38 Bottom-up eamination of SRAM circuits (5) Address decoded to select one row of bitslice for read/write :4 Decoder bitline bitline 1 bitline 0 Address wordline 3 wordline wordline 1 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 38

39 Bottom-up eamination of SRAM circuits (6) To increase word size, add bitslices. :4 Decoder bitline bitline 1 bitline 0 Address wordline 3 wordline wordline 1 wordline 0 Data Data 1 Data 0 Copyright 007 Elsevier 39

40 Coincident cell selection Decode address into both row and column select signals How many words in this RAM? How many bits per word? 40

41 Coincident cell selection w. larger words How many words in this RAM? How many bits per word? 41

42 Coincident cell selection saves decode logic STORAGE COLUMNS 15-bit address DECODER 3,768 row selects... 3K words STORAGE COLUMN 9-bits of address DECODER #1 51 row selects words 1 byte... 1 byte...64 column selects... DECODER # (3,800 gates) (608 gates) 6-bits of address 4

43 Multi-chip memories If you need a larger memory than any available chip Wire multiple RAM chips together to wor in concert as one large memory 43

44 Memory Timing: Write eample Even though memory not on the cloc, timing still an issue: inputs are on the cloc must first be properly enabled for reading or writing before data is transferred Appropriate address chosen Appropriate segment enabled Appropriate read/write configuration set Data valid: period during which writing must be performed Time Address Mem enable Read/Write Data Valid Write eample 44

45 Memory Timing: Read eample Even though memory not on the cloc, timing still an issue: inputs are on the cloc must first be properly enabled for reading or writing before data is transferred Appropriate address chosen Appropriate segment enabled Appropriate read/write configuration set Data valid: period during which writing must be performed Time Address Mem enable Read/Write Data Valid Read eample CPU to Mem communication delay 45

46 Programmable Logic Devices Programmable logic devices (PLDs) Structured lie memories Used to implement combinational logic X on array logic means wire connected to logic gate (e.g. above) Connections can be either permanent (e.g., fuse, mas) or not (e.g., Flash) 46

47 General PLD architecture... Fied AND, programmable OR = Programmable ROM (PROM) Programmable AND, fied OR = Programmable Array Logic (PAL) Programmable AND, programmable OR = Programmable Logic Array (PLA) 47

48 Programmable ROM (PROM) A B C D m0 m1... m14 m15 E F G H Fied (X) AND, programmable (X) OR 48

49 Programmable Array Logic (PAL) A B C D... E F Programmable (X) AND, fied (X) OR 49

50 Programmable Logic Array (PLA) A B C D... E F G H Programmable (X) AND, programmable (X) OR 50