Dec 2010VLSI Interconnects 1 Instructed by Shmuel Wimer Eng. School, Bar-Ilan University Credits: David Harris Harvey Mudd College (Some material copied/taken/adapted from Harris’ lecture notes)

Dec 2010VLSI Interconnects 2 Outline  Introduction  Wire Resistance  Wire Capacitance  Wire RC Delay  Crosstalk  Wire Engineering  Repeaters  Scaling

Dec 2010VLSI Interconnects 3 Introduction  Chips are mostly made of wires called interconnect  Alternating layers run orthogonally

Dec 2010VLSI Interconnects 4  Transistors are little things under the wires  Many layers of wires

Dec 2010VLSI Interconnects 5  Wires are as important as transistors –Speed –Power –Noise

Dec 2010VLSI Interconnects 6 Wire Geometry  Pitch = w + s  Aspect ratio: AR = t/w –Old processes had AR << 1 –Modern processes have AR  2 Pack in many skinny wires l ws t h

Dec 2010VLSI Interconnects 7 Layer Stack  Modern processes use metal layers  Example: Intel 180 nm process  M1: thin, narrow (< 3 ) –High density cells  M2-M4: thicker –For longer wires  M5-M6: thickest –For V DD, GND, clk Substrate LayerT(nm) AR W (nm) S

Dec 2010VLSI Interconnects 8 Wire Resistance   = resistivity (  *m)  R  = sheet resistance (  /  ) –  is a dimensionless unit(!)  Count number of squares –R = R  * (# of squares) l w t 1 Rectangular Block R = R (L/W)  4 Rectangular Blocks R = R (2L/2W)  = R (L/W)  t l ww l

Dec 2010VLSI Interconnects 9 Choice of Metals  Until 180 nm generation, most wires were aluminum  Modern processes use copper –Cu atoms diffuse into silicon and damage FETs –Must be surrounded by a diffusion barrier Metal Bulk resistivity (  *cm) Silver (Ag)1.6 Copper (Cu)1.7 Gold (Au)2.2 Aluminum (Al)2.8 Tungsten (W)5.3 Molybdenum (Mo)5.3

Dec 2010VLSI Interconnects 10 Sheet Resistance  Typical sheet resistances in 180 nm process Layer Sheet Resistance (  /  ) Diffusion (silicided)3-10 Diffusion (no silicide) Polysilicon (silicided)3-10 Polysilicon (no silicide) Metal10.08 Metal20.05 Metal30.05 Metal40.03 Metal50.02 Metal60.02

Dec 2010VLSI Interconnects 11 Contacts Resistance  Contacts and vias also have 2-20   Use many contacts for lower R –Many small contacts for current crowding around periphery

Dec 2010VLSI Interconnects 12 Wire Capacitance  Wire has capacitance per unit length –To neighbors –To layers above and below  C total = C top + C bot + 2C adj layer n+1 layer n layer n-1 C adj C top C bot ws t h 1 h 2

Dec 2010VLSI Interconnects 13 Capacitance Trends  Parallel plate equation: C =  A/d –Wires are not parallel plates, but obey trends –Increasing area (W, t) increases capacitance –Increasing distance (s, h) decreases capacitance  Dielectric constant –  = k  0   0 = 8.85 x F/cm  k = 3.9 for SiO 2  Processes are starting to use low-k dielectrics –k  3 (or less) as dielectrics use air pockets

Dec 2010VLSI Interconnects 14 M2 Capacitance Data  Typical wires have ~ 0.2 fF/  m –Compare to 2 fF/  m for gate capacitance  Capacitance increases with metal planes above and below  Capacitance decreases with spacing

Dec 2010VLSI Interconnects 15 Diffusion & Polysilicon  Diffusion capacitance is very high (about 2 fF/  m) –Comparable to gate capacitance –Diffusion also has high resistance –Avoid using diffusion (and polysilicon) runners for wires!  Polysilicon has lower C but high R –Use for transistor gates –Occasionally for very short wires between gates

Dec 2010VLSI Interconnects 16 Lumped Element Models  Wires are a distributed system –Approximate with lumped element models  3-segment  -model is accurate to 3% in simulation

Dec 2010VLSI Interconnects 17

Dec 2010VLSI Interconnects 18

Dec 2010VLSI Interconnects 19 RC Example  Metal2 wire in 180 nm process –5 mm long –0.32  m wide  Construct a 3-segment  -model –R  = 0.05  /  => R = 781  –C permicron = 0.2 fF/  m => C = 1 pF 260  167 fF 260  167 fF 260  167 fF

Dec 2010VLSI Interconnects 20

Dec 2010VLSI Interconnects 21 Wire RC Delay Example  Estimate the delay of a 10x inverter driving a 2x inverter at the end of the 5mm wire from the previous example –R = 2.5 k  *  m for gates –Unit inverter: 0.36  m nMOS, 0.72  m pMOS –t pd = 1.1 ns 781  500 fF Wire 4 fF LoadDriver 690 

Dec 2010VLSI Interconnects 22 Crosstalk  A capacitor does not like to change its voltage instantaneously  A wire has high capacitance to its neighbor. –When the neighbor switches from 1→ 0 or 0→ 1, the wire tends to switch too. –Called capacitive coupling or crosstalk  Crosstalk effects –Noise on non switching wires –Increased delay on switching wires

Dec 2010VLSI Interconnects 23 Miller Effect  Assume layers above and below on average are quiet –Second terminal of capacitor can be ignored –Model as C gnd = C top + C bot  Effective C adj depends on behavior of neighbors –Miller effect B VV C eff(A) MCF ConstantV DD C gnd + C adj 1 Switching with A 0C gnd 0 Switching opposite A 2V DD C gnd + 2 C adj 2 AB C adj C gnd C Delay and power increase

Dec 2010VLSI Interconnects 24 Crosstalk Noise  Crosstalk causes noise on non switching wires  If victim is floating: –model as capacitive voltage divider C adj C gnd-v Aggressor Victim  V aggressor  V victim

Dec 2010VLSI Interconnects 25 Driven Victims  Usually victim is driven by a gate that fights noise –Noise depends on relative resistances –Victim driver is in linear region, agg. in saturation –If sizes are same, R aggressor = 2-4 x R victim C adj C gnd-v Aggressor Victim  V aggressor  V victim R aggressor R victim C gnd-a

Dec 2010VLSI Interconnects 26 Coupling Waveforms  Simulated coupling for C adj = C victim Aggressor Victim (undriven): 50% Victim (half size driver): 16% Victim (equal size driver): 8% Victim (double size driver): 4% t (ps)

Dec 2010VLSI Interconnects 27 Noise Implications  So what if we have noise?  If the noise is less than the noise margin, nothing happens  Static CMOS logic will eventually settle to correct output even if disturbed by large noise spikes –But glitches cause extra delay –Also cause extra power from false transitions  Dynamic logic never recovers from glitches  Memories and other sensitive circuits also can produce the wrong answer

Dec 2010VLSI Interconnects 28 Wire Engineering

Dec 2010VLSI Interconnects 29  Goal: achieve delay, area, power goals with acceptable noise  Degrees of freedom: –Width –Spacing –Layer –Shielding vdda 0 a 1 gnda 2 vddb 0 a 1 a 2 b 2 a 0 a 1 gnda 2 a 3 vddgnda 0 b 1

Dec 2010VLSI Interconnects 30 Repeaters  R and C are proportional to l  RC delay is proportional to l 2 –Unacceptably great for long wires  Break long wires into N shorter segments –Drive each one with an inverter or buffer Wire Length: l DriverReceiver l/N Driver Segment Repeater l/N Repeater l/N ReceiverRepeater N Segments

Dec 2010VLSI Interconnects 31 Repeater Design  How many repeaters should we use?  How large should each one be?  Equivalent Circuit –Wire length l Wire Capacitance C w *l, Resistance R w *l –Inverter width W (nMOS = W, pMOS = 2W) Gate Capacitance C’*W, Resistance R/W R/W C'WC w l /2NC w l R w l /N

Dec 2010VLSI Interconnects 32

Dec 2010VLSI Interconnects 33 w/o repeater k=1 Showing that repeater insertion pays small

Dec 2010VLSI Interconnects 34 Under what condition repeater insertion should take place?

Dec 2010VLSI Interconnects 35

Dec 2010VLSI Interconnects 36

Dec 2010VLSI Interconnects 37

Dec 2010VLSI Interconnects 38 Cascaded driver with non-repeated interconnect Non-repeated interconnect The method is useful when R tr is dominant and C int is large

39 v u, q Optimal Buffer Insertion ? ? ? ? ? ? ?

40 Delay Model R3R3 C1C1 R2R2 R6R6 R5R5 R4R4 R7R7 C6C6 C5C5 C4C4 C7C7 R1R1 C2C2 C3C

41 Bottom-Up Solution RKRK CKCK K (T’ K, L’ K ) RMRM CMCM M sub-tree (T M, L M ) RNRN CNCN N sub-tree (T N, L N ) (T K, L K )

Dec 2010VLSI Interconnects 42 Scaling

Dec 2010VLSI Interconnects 43

Dec 2010VLSI Interconnects 44

Dec 2010VLSI Interconnects 45

Dec 2010VLSI Interconnects 46 Scaling