Title: CS184a:%20Computer%20Architecture%20(Structure%20and%20Organization)
1CS184aComputer Architecture(Structure and
Organization)
- Day 6 January 22, 2003
- VLSI Scaling
2Today
- VLSI Scaling Rules
- Effects
- Historical/predicted scaling
- Variations (cheating)
- Limits
3Why Care?
- In this game, we must be able to predict the
future - Rapid technology advance
- Reason about changes and trends
- re-evaluate prior solutions given technology at
time X.
4Why Care
- Cannot compare against what competitor does today
- but what they can do at time you can ship
- Careful not to fall off curve
- lose out to someone who can stay on curve
5Scaling
- Premise features scale uniformly
- everything gets better in a predictable manner
- Parameters
- l (lambda) -- Mead and Conway (class)
- S -- Bohr
- 1/k -- Dennard
6Feature Size
l is half the minimum feature size in a
VLSI process minimum feature usually
channel width
7Scaling
- Channel Length (L)
- Channel Width (W)
- Oxide Thickness (Tox)
- Doping (Na)
- Voltage (V)
8Scaling
- Channel Length (L) l
- Channel Width (W) l
- Oxide Thickness (Tox) l
- Doping (Na) 1/l
- Voltage (V) l
9Effects?
- Area
- Capacitance
- Resistance
- Threshold (Vth)
- Current (Id)
- Gate Delay (tgd)
- Wire Delay (twire)
- Power
10Area
- 0.35mm ? 0.25mm
- 50 area
- 2x capacity same area
11Area Perspective
2000 tech. 18mm?18mm 0.18mm 60G l2
12Capacity Scaling from Intel
13Capacitance
- Capacitance per unit area
- Cox eSiO2/Tox
- Tox? Tox/k
- Cox ? k Cox
14Capacitance
- Gate Capacitance
- Cgate ACox
- A ? A/k2
- Cox ? k Cox
- Cgate ? Cgate /k
15Threshold Voltage
16Threshold Voltage
17Current
- Saturation Current
- Id(mCOX/2)(W/L)(Vgs-VTH)2
- VgsV? V /k
- VTH? VTH /k
- W? W/k
- Cox ? k Cox
- Id? Id/k
18Gate Delay
- tgdQ/I(CV)/I
- V? V /k
- Id ? Id/k
- C ? C /k
- tgd ? tgd /k
19Resistance
- RrL/(Wt)
- W? W/k
- L, t similar
- R ? k R
20Wire Delay
- twireR?C
- R -gt k R
- C -gt C /k
- twire -gt twire
- assuming (logical) wire lengths remain
constant... - Assume short wire or buffered wire
- (unbuffered wire ultimately scales as length
squared)
21Power Dissipation (Static)
- Resistive Power
- PVI
- V? V /k
- Id? Id/k
- P? P /k2
22Power Dissipation (Dynamic)
- Capacitive (Dis)charging
- P(1/2)CV2f
- V? V /k
- C ? C /k
- P? P/k3
- Increase Frequency?
- f ? kf ?
- P ? P/k2
23Effects?
- Area 1/k2
- Capacitance 1/k
- Resistance k
- Threshold (Vth) 1/k
- Current (Id) 1/k
- Gate Delay (tgd) 1/k
- Wire Delay (twire) 1
- Power 1/k2? 1/k3
24ITRS Roadmap
- Semiconductor Industry rides this scaling curve
- Try to predict where industry going
- (requirementsself fulfilling prophecy)
- http//public.itrs.net
25MOS Transistor Scaling(1974 to present)
from Andrew Kahng
Source 2001 ITRS - Exec. Summary, ORTC Figure
26Half Pitch ( Pitch/2) Definition
from Andrew Kahng
Source 2001 ITRS - Exec. Summary, ORTC Figure
27Scaling Calculator Node Cycle Time
from Andrew Kahng
Source 2001 ITRS - Exec. Summary, ORTC Figure
28from Andrew Kahng
Source 2001 ITRS - Exec. Summary, ORTC Figure
29Delays?
- If delays in gates/switching?
- If delays in interconnect?
- Logical interconnect lengths?
30Delays?
- If delays in gates/switching?
- Delay reduce with 1/k l
31Delays
- Logical capacities growing
- Wirelengths?
- No locallity?k (slower!)
- Rents Rule
- L ? n(p-0.5)
- pgt0.5
32Compute Density
- Density compute / (Area Time)
- k3gtcompute density scalinggtk
- k3 gates dominate, plt0.5
- k2 moderate p, good fraction of gate delay
- p from Rents Rule again more on Day12
- k large p (wires dominate area and delay)
33Power Density
- P-gt P /k2 (static, or increase frequency)
- P-gt P/k3 (dynamic, same freq.)
- A -gt A/k2
- P/A ? P/A or P/kA
34Cheating
- Dont like some of the implications
- High resistance wires
- Higher capacitance
- Need for more wiring
- Not scale speed fast enough
35Improving Resistance
- RrL/(Wt)
- W? W/k
- L, t similar
- R ? k R
- Dont scale t quite as fast.
- Decrease r (copper)
36Improving Capacitance
- Capacitance per unit area
- Cox eSiO2/Tox
- Tox? Tox/k
- Cox ? k Cox
Reduce Dielectric Constant e
37Wire Layers More Wiring
38Typical chip cross-section illustrating hierarchic
al scaling methodology
Passivation
Dielectric
Wire
Etch Stop Layer
Via
Global (up to 5)
Dielectric Capping Layer
Copper Conductor with Barrier/Nucleation Layer
Intermediate (up to 4)
Local (2)
Pre Metal Dielectric
Tungsten Contact Plug
from Andrew Kahng
39Improving Gate Delay
- tgdQ/I(CV)/I
- V? V /k
- Id(mCOX/2)(W/L)(Vgs-VTH)2
- Id ? Id/k
- C ? C /k
- tgd ? tgd /k
Dont scale V V?V I?kI tgd ? tgd /k2
40But Power Dissipation (Dynamic)
- Capacitive (Dis)charging
- P(1/2)CV2f
- V? V /k
- C ? C /k
- P? P/k3
- Increase Frequency?
- f ? kf ?
- P ? P/k2
If not scale V, power dissipation not scale.
41AndPower Density
- P? P (increase frequency)
- P?gt P/k (dynamic, same freq.)
- A ? A/k2
- P/A ? kP/A or k2P/A
- Power Density Increases
42Physical Limits
43Physical Limits
- Depended on
- bulk effects
- doping
- current (many electrons)
- mean free path in conductor
- localized to conductors
- Eventually
- single electrons, atoms
- distances close enough to allow tunneling
44What Is A Red Brick ?
- Red Brick ITRS Technology Requirement with no
known solution - Alternate definition Red Brick something
that REQUIRES billions of dollars in RD
investment
from Andrew Kahng
45The Red Brick Wall - 2001 ITRS vs 1999
Source Semiconductor International -
http//www.e-insite.net/semiconductor/index.asp?la
youtarticlearticleIdCA187876
from Andrew Kahng
46Finishing Up...
47Big IdeasMSB Ideas
- Moderately predictable VLSI Scaling
- unprecedented capacities/capability growth for
engineered systems - change
- be prepared to exploit
- account for in comparing across time
48Big IdeasMSB-1 Ideas
- Uniform scaling reasonably accurate for past
couple of decades - Area increase k2
- Real capacity maybe a little less?
- Gate delay decreases (1/k)
- Wire delay not decrease, maybe increase
- Overall delay decrease less than (1/k)