Prof. John Nestor ECE Department Lafayette College Easton, Pennsylvania ECE VLSI Circuit Design Lecture 23 - Subsystem Design Spring 2007

ECE 425 Spring 2007Lecture 23 - Subsystem Design2 Announcements  Reading  Wolf:  These notes drawn in part from handouts by  J. Rabaey, Digital Integrated Circuits, © Prentice-Hall 1995.

ECE 425 Spring 2007Lecture 23 - Subsystem Design3 Where we Are:  Last Time:  More about Project Pad Frame & Comparator  Streaming Video: Two Talks on Chip-Level Design  Today:  Custom Subsystem Design General approach Shifters Adders

ECE 425 Spring 2007Lecture 23 - Subsystem Design4 Subsystem Design  General Techniques   Goals  Pipelining  Datapath Design  Common Subsystems  Shifters  Adders  ALUs  Multipliers  Memories  Structure Logic

ECE 425 Spring 2007Lecture 23 - Subsystem Design5 Subsystem Design (Ch. 6)  Goals of Custom Subsystem Design  Maximize performance  Minimize area  Fit together with other subsystems  Key idea: optimize across levels of abstractions  Layout  Circuit  Logic  Register-Transfer and Higher

ECE 425 Spring 2007Lecture 23 - Subsystem Design6 Optimizing for Peformance and/or Area  Layout level  Microscopic changes: move wires, change wire sizing add vias, reduce source/drain cap, etc.  Macroscopic changes: cell placement, design of hierarchy  Circuit level  Transistor sizing  Advanced circuits (e.g., dynamic logic)

ECE 425 Spring 2007Lecture 23 - Subsystem Design7 Optimizing for Peformance and/or Area (cont'd)  Logic level  Use specialized designs (e.g. shifters, ALUs, etc.)  Flatten to reduce delay  Restructure  Register-transfer level (and above)  Place latches/flip flops to maximize performance (retiming)  Encode FSMs to minimize area/delay  Perform computations in parallel with extra hardware if cost permits  Pipeline logic to increase performance

ECE 425 Spring 2007Lecture 23 - Subsystem Design8 Pipelining  Key idea: Partition combinational function with latches / flip flops  Each partition is called a stage  Time between each result: one clock period  Latency: number of clock cycles before result appears (== number of stages)

ECE 425 Spring 2007Lecture 23 - Subsystem Design9 Example - Before Pipelining  Comb. logic delay  t p =80ns*  Latch/Flip-Flop setup  t su =5ns  Clock Period  t clk =85ns * archiac TTL timing values - divide by approx. 100 for VLSI!

ECE 425 Spring 2007Lecture 23 - Subsystem Design10 Example - After Pipelineing  Comb. logic delay  t p1 =t p2 =40ns  Latch setup  t su =5ns  Clock Period  t clk =45ns  Latency: 2 cycles

ECE 425 Spring 2007Lecture 23 - Subsystem Design11 Pipelining Comments  Impact on performance  Increases operations per unit time  Increases latency  Added overhead due to register setup times  Design concerns / limits  Balance stage delays for best performance  Structure of logic may limit number of stages

ECE 425 Spring 2007Lecture 23 - Subsystem Design12 Effect of Adding Pipeline Stages

ECE 425 Spring 2007Lecture 23 - Subsystem Design13 Custom Datapath Design  Goal: create a tight design of several elements  Arithmetic / Logic Functions, Shifters  Storage: Registers, Register Files  Interconnect: wires, buses

ECE 425 Spring 2007Lecture 23 - Subsystem Design14 Datapath Pysical Design  Bit-sliced layout of each component  Connection by abutment  "Pitch-matched" connections  Designed using "wiring plan" Wiring Plan

ECE 425 Spring 2007Lecture 23 - Subsystem Design15 Bus Design in datapaths  Key idea: replace multiplexers with distributed drivers for long connections  Pseudo-nmos NOR: Fig 6-8, p. 318 high power simple design  Precharged: Fig 6-9, p. 319 lower power more complex design  In either case, careful circuit design and interconnect modeling is essential Pseudo-nmos bus Precharged bus

ECE 425 Spring 2007Lecture 23 - Subsystem Design16 Ancient Example: the Motorola 68K

ECE 425 Spring 2007Lecture 23 - Subsystem Design17 Subsystem Design  General Techniques  Goals  Pipelining  Datapath Design  Common Subsystems   Shifters  Adders  ALUs  Multipliers  Memories  Structure Logic

ECE 425 Spring 2007Lecture 23 - Subsystem Design18 Shifter Design  Why shift?  Arithmetic operations  Floating-point  Bit field extraction  Shift Register - one shift per clock cycle  Hardware shifters - implement as comb. logic  Single-bit shifters  Barrel shifters  Logarithmic shifters

ECE 425 Spring 2007Lecture 23 - Subsystem Design19 Single-Bit Shifter  Essentially a MUX made from pass transistors Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall.

ECE 425 Spring 2007Lecture 23 - Subsystem Design Barrel Shifter  pass transistors connect input bit to chosen output  regular layout  each signal flows through only one trans. gate  area dominated by pitch of metal wires

ECE 425 Spring 2007Lecture 23 - Subsystem Design21 4 X 4 Barrel Shifter - Layout Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall Width barrel ~ 2 p m M

ECE 425 Spring 2007Lecture 23 - Subsystem Design22 Logarithmic Shifter  Combine shifts of powers-of-two Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design23 Logarithmic Shifter - Layout Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design24 Adder Design  Review: Full Adder  Sum:s i = a i XOR b i XOR c i  Carry:c i+1 = a i *b i + a i *c i + b i *c i AiAi BiBi SiSi CiCi C i+1 AiAi BiBi CiCi SiSi

ECE 425 Spring 2007Lecture 23 - Subsystem Design25 Adder Design (cont'd)  Ripple: constructed from n full adders  Compact, but delay proportional to n  May be tolerable when n=8, BUT  What about n=32? Potential worst cases: A 0 or B 0 to S 31 A 0 or B 0 to C 32 A0A0 B0B0 S0S0 C0C0 C1C1 A1A1 B1B1 S1S1 C1C1 C2C2 A2A2 B2B2 S2S2 C2C2 C3C3 A3A3 B3B3 S3S3 C3C3 C4C4 0

ECE 425 Spring 2007Lecture 23 - Subsystem Design26 Full Adder - Static CMOS Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design27 Inversion Property Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design28 Inversion Adder Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design29 Mirror Adder: A Better Structure Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design30 Mirror adder notes The NMOS and PMOS chains are completely symmetrical. This guarantees identical rising and falling transitions if the NMOS and PMOS devices are properly sized. A maximum of two series transistors can be observed in the carry-generation circuitry. When laying out the cell, the most critical issue is the minimization of the capacitance at node C o. The reduction of the diffusion capacitances is particularly important. The capacitance at node C o is composed of four diffusion capacitances, two internal gate capacitances, and six gate capacitances in the connecting adder cell. The transistors connected to C i are placed closest to the output. Only the transistors in the carry stage have to be optimized for optimal speed. All transistors in the sum stage can be minimal size. Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design31 Dynamic Adder - np-CMOS Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall 17 transistors

ECE 425 Spring 2007Lecture 23 - Subsystem Design32 Layout - Dynamic np-CMOS adder Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design33 Speeding up Carry - Carry Lookahead  Key idea: trade off delay, amount of logic used  Benefit: Faster addition  Cost: much more logic  Define two signals for each adder stage:  Generateg i = a i *b i  Propagatep i = a i + b i  Why use these names?  Adder i will always generate a carry if a i, b i both true  A i will propagate a carry input if either or both a i, b i both true A0A0 B0B0 S0S0 C0C0 C1C1 11 X

ECE 425 Spring 2007Lecture 23 - Subsystem Design34 Carry Lookahead (cont’d)  Now rewrite carry output as function of a i,b i,p i,g i  Original eqn: c i+1 = a i *b i + a i *c i + b i *c i  New eqn:c i+1 = g i + p i *c i  "Flatten" carry function in terms of g i, p i c 1 = g 0 + p 0 *c 0 c 2 = g 1 + p 1 *c 1 = g 1 + p 1 *(g 0 + p 0 *g 0 ) = g 1 + p 1 *g 0 + p 1 *p 0 *c 0 c 3 = g 2 + p 2 *g 1 + p 2 *p 1 *g 0 + p 3 *p 2 *p 1 *c 0 c 4 = g 3 + p 3 *g 2 + p 3 *p 2 *g 1 + p 3 *p 2 *p 1 *g 0 + p 3 *p 2 *p 1 *p 1 *c 0  Add carry lookahead logic that computes c 1 -c 4 in terms of p 0 - p 3 and g 0 -g 3

ECE 425 Spring 2007Lecture 23 - Subsystem Design35 Logarithmic Lookahead: Brent-Kung Adder Source: J. Rabaey, Digital Integrated Circuits © 1995 Prentice-Hall

ECE 425 Spring 2007Lecture 23 - Subsystem Design36 Coming Up:  More about adders  ALUs  Memories  Structure Logic