Introduction. Summary. Why computer architecture? Technology trends Cost issues

Transcription

1 Introduction 1 Summary Why computer architecture? Technology trends Cost issues 2 1

2 Computer architecture? Computer Architecture refers to the attributes of a system visible to a programmer (that have a direct impact on the logical execution of a program) Architectural attributes: Instruction set Data type representations I/O mechanisms Memory addressing techniques 3 Computer architecture (2)? Computer organization refers to the operational units and their interconnections that realize the architectural specifications Organizational attributes: hardware details transparent to the programmer: control signals interfaces between the computer and peripherals memory technology Often called micro-architecture 4 2

3 Architecture vs. organization Example: Architectural issue: Does the computer support the multiplication instruction? Organizational issue: Is multiplication imlemented by a special multiply unit or by a add+accumulate? 5 Architecture vs. organization (2) Architecture (MIPS): Instruction Formats op rs rt rdshamtfunct op rs rt offset op address Registers 32 bits R0 R1 R2 R30 R31 32 General Purpose Registers PC = 0x C 0x x x x C 0x x x x C 32 0xfffffffc 0xfffffffc Memory (Max. 4GB) 32 bits 6 3

4 Architecture vs. organization (3) Pentium 4 7 Architecture vs. organization (4) 8 4

5 Architecture vs. organization (3) Often: computer architecture = Instruction Set Architecture (ISA) The true Hardware-Software interface The most important abstraction of computer design Application Application Programs Operating System Software Compiler Processor I/O System Instruction Set Architecture Interface between SW & HW Logic - gates, state machines, etc. Circuit - transistors, etc. Hardware Layout - mask patterns, etc. 9 Computer architecture Why is it useful? Micro-architecture Not likely you are going to be microprocessor architects Software Designers: How does software really get executed? What about software design effects performance Hardware Designers: What should go in hardware, what should go in software? 10 5

6 Computer architecture In practice, we will cover: Computer architecture ISA Computer organization Processor structure Processor execution mechanism Pipelining/Instruction-set parallelism Memory hierarchy principles I/O interface Specialization of architecture/organization Non-conventional architectures Multimedia extensions (some) 11 Computer architecture Interaction with many disciplines 12 6

7 Objective Before getting into the details, we must address: 1. Technology trends 2. Cost issues 3. Performance issues 13 Technology trends 14 7

8 A brief computer history 1940s-50s - Vacuum Tubes 1950s-60s - Discrete Transistors 1960s-70s - Discrete ICs (e.g., TTL) 1970s-present - LSI and VLSI microprocessors 15 A brief computer history (2) ENIAC s (Vacuum Tubes) IBM s (Transistors) 16 8

9 A brief computer history (3) Intel s (First Microprocessor) DEC VAX 11/ s (Discrete IC s) 17 A brief computer history (4) MOS Technology 6502 Apple II Computer 1970s 18 9

10 A brief computer history (5) Intel 8088 (LSI Microprocessor) Original IBM PC 1980s 19 A brief computer history (6) Pentium III 28M transistors / 733MHz-1Gz / 13-26W L=0.25µm shrunk to L=0.18µm 1990s PowerPC 7400 (G4) 6.5M transistors / 450MHz / 8-10W L=0.15µm 20 10

11 A brief computer history (5) Pentium 4 42M transistors / GHz / 49-55W L=0.18µm...today Pentium 4 Northwood 55M transistors / 2-2.5GHz L=0.13µm 21 A brief computer history (6)...the next step

12 Building blocks Current dominant technology (for digital circuits!): Complementary MOS (CMOS) Other technologies: Bipolar (e.g., TTL) Bi-CMOS - hybrid Bipolar, CMOS GaAs - Gallium Arsenide (for high speed) SiGe - Silicon Germanium (for high speed, RF) Used for different targets 23 CMOS transistor (n-type) Basically, a switch G D S G=1 G=0 Source SiO2 Insulator W L n+ n+ channel p substrate Gate Drain Technology characterized by the value of L (transistor length) Feature size Today L=0.13 µm (!)...trend : L=130nm 2003: L=90nm 2005: L=65nm? 24 12

13 CMOS scaling Features size scales down quite fast 2001 forecast Year µm And even faster than forecasted 1997 forecast Year µm An example: memory chip IBM J. R&D, Jan/Mar 95 Evolution from Mb 256 Mb uses cell with area 0.6 µm2 Cell Layouts 4 Mb Cell Structure 4Mb 16Mb 64Mb 256Mb 26 13

14 Moore s Law In 1965, Gordon Moore predicted that transistors would continue to shrink, allowing: Doubled transistor density every months Doubled performance every months History has proven Moore right But, is the end in sight? Physical limitations: quantum effects create fundamental limits as approach atomic scale Economic limitations 27 Orders of magnitude 1 cm 1 mm 0.1 mm 10µm 1 µm 0.1 µm 10 nm 1 nm 1 Å Chip size (1 cm) Diameter of Human Hair (25 µm) 1996 devices (0.35 µm) 2007 devices (0.1 µm) Silicon atom radius (1.17 Å) Deep UV Wavelength (248 nm) X-ray Wavelength (0.6 nm) 28 14

15 Microprocessor Trends (1) 100,000,000 Transistors 10,000,000 1,000, ,000 Moore s Law i80286 i80386 Pentium i80486 Pentium 4: 55 million Alpha 21264: 15 million Pentium Pro: 5.5 million PowerPC 620: 6.9 million Alpha 21164: 9.3 million Sparc Ultra: 5.2 million 10,000 1,000 i4004 i8080 i Log scale! Year Current trend: doubles every 18 month => +54% /year 29 Intel family Microprocessor Trends (2) Year Chip L transistors µm 2.3K µm 6.0K µm 29K µm 134K µm 275K µm 1.2M 1993 Pentium 0.8µm 3.1M 1995 Pentium Pro 0.6µm 15.5M 1999 Mobile PII 0.25µm Pentium µm 42M 2002 Pentium 4 (N) 0.13µm 55M Source:

16 1200 Relative speed Microprocessor Trends (3) DEC Alpha 21264/ DEC HP IBM MIPS MIPS AXP/ Sun-4/ 260 M 2000 M/ 120 RS/ 9000/ DEC Alpha 5/500 DEC Alpha 5/300 DEC Alpha 4/266 IBM POWER Current trend: doubles every 18 month => +54% /year 31 DRAM Memory Trends 1,000,000, ,000,000 10,000,000 1,000, , size year size(mb) cyc time ns ns ns x 4 0.5x 165 ns ns ns ns Year Log scale! Size trend: ~doubles every 18 month => +54% /year Speed: ~ doubles every 10 years! => +9% /year 32 16

17 Memory-processor gap It is not just a matter of faster processors!!!! but this only a part of the picture 33 Summary - Technology Trends Processor Logic capacity Clock frequency Cost per function Memory DRAM capacity: (4x every 3 years) Speed: Cost per bit: increases ~ 50% per year increases ~ 20% per year decreases ~20% per year increases ~ 60% per year increases ~ 10% per year decreases ~25% per year 34 17

18 Summary - Technology Trends SIA Roadmap (2001) Note: 50nm = 5*10-9 m Silicon atom O(10-10 m) Message: Scaling improves logic speed & density Scaling increases power! 35 Power constraints Projecting power consumption Will soon get to KW? Yes, if we do not use power reduction solutions? It is not just a matter of faster processors! (again) 36 18

19 Wires vs. gates Scaling applies to transistors as well as to wires Wire scaling means OLD NEW H H L W W Capacitance does not scale well Ground capacitance scales Lateral capacitance increases Overall capacitance increase w.r.t. gate capacitances Higher capacitance C Longer delays (T=RC) Higher power consumption (P=CV 2 ) L 37 Wire vs. gate delay It is not just a matter of faster logic! 38 19

20 Common belief Cost issues Performance is top priority Power almost top priority Area (size) is irrelevant Real estate on a chip is not a problem However, chip cost is directly related to area 39 VLSI Processing 40 20

21 Integrated Circuits Costs Die cost + Testing cost + Packaging cost IC cost = Final test yield Wafer cost Die cost = Dies per Wafer * Die yield die per wafer = π * (Wafer_diam/2) 2 π * Wafer_diam - Die Area (2 * Die Area) 1/2 - test dies Die Yield = Wafer yield * 1 + { α Defects_per_unit_area * Die_Area Die Cost roughly die_area 4 α } 41 Examples Chip Metal Line Wafer Defect Area Dies/ Yield Die Cost layers width cost /cm 2 mm 2 wafer 386DX $ % $4 486DX $ % $12 PowerPC $ % $53 HP PA $ % $73 DEC Alpha $ % $149 SuperSPARC $ % $272 Pentium $ % $417 From "Estimating IC Manufacturing Costs, Linley Gwennap, Microprocessor Report, August 2, 1993, p

22 Logic vs. I/O Scaling affects speed of logic Only marginally affects speed of devices! Impact of scaling on computation: application Web surfing It is not just a matter of faster processors! (again) 43 In summary Overall system performance depends on: Processor speed Memory architecture I/O system Performance are constrained by: Power consumption Area (translates into cost) Do not neglect hidden costs Technological issues Wires (delay and power) cost dominates gate cost Small feature sizes imply increasing parasitic effects Impact of software 44 22