Title: Integrated Regulation for Energy-Efficient Digital Circuits
1Integrated Regulation for Energy-Efficient
Digital Circuits
- Elad Alon1 and Mark Horowitz2
1UC Berkeley 2Stanford University
2Scaling and Supply Impedance
- CMOS scaling led to lower supply voltages and
constant (or increasing) power consumption
Impedance Requirements of High-Performance
Processors
- Forces drastic drop in supply impedance
- Vdd ?, Idd ? ? Zrequired ??
- Todays chips
- Zrequired 1 mO!
- Hard to achieve across a broad frequency range
Technology (µm)
3Power Distribution and Regulation
- Significant resources spent to meet impedance
requirement - Active regulation can be used to reduce impedance
- E.g., Active/switched decoupling
- But, these regulators increase total power
- Prevents adoption in todays power-limited chips
4Power-Neutral Regulation
Vdd
- Gate delay depends on Vdd
- So Vdd needs to be greater than some Vmin
- Supply variations force higher nominal voltage
- Causes extra power dissipation
Vnom
Vmin
- Goal make regulator power less than power
recovered from lower noise
5Outline
- Regulator Topology
- Regulator Design
- Experimental Verification
- Conclusions
6Linear Regulators
Series Regulator
Chip
-
Vdd
Load
Vreg
Vref
-
Shunt Regulator
Chip
-
Load
Vdd
Vreg
Vref
-
7Series Regulator Efficiency
Vdd
Vref
Vdd
Noise
Vreg
Vdrop
Vreg
Load
- Clearly wont meet efficiency goal
- Regulator doesnt really change noise on Vdd
- So still need same margin
- But added an extra Vdrop from variable resistor
8Shunt Regulator Efficiency
Itotal
max(Inoise)
Iload
Ishunt
max(Inoise)
0
- Regulator can only pull current out of supply
- Need to burn significant static current to
counter noise in both directions - Again, clearly inefficient
- Need to allow shunt to deliver energy to the load
- Not just dissipate it
9Push-Pull Shunt Regulator
Vshunt
Ipush
Iload
Vreg
-
Vref
Ipush
Ipull
Load
-
Ipull
0
- Use an additional, shunt supply to push current
into Vreg - Regulator capable of countering large variations
- But regulator loss set mostly by average
variation - Similar to Active Clamp for board VRMs
- Build on previous work to improve on-die impedance
A.M. Wu and S.R. Sanders, An Active Clamp
Circuit for Voltage Regulation Module (VRM)
Applications, Transactions on Power Electronics.
Sept. 2001.
10Outline
- Regulator Topology
- Regulator Design
- Shunt Supply Network
- Minimizing Static Power
- Experimental Verification
- Conclusions
11Regulator Design Challenges
- Vshunt is not free
- Takes resources away from main supply
- Increases loss
- Need to minimize quiescent output current
- Otherwise regulator too inefficient
- Need GHz bandwidth feedback path
- With minimum feedback circuit power
- Output stage in particular is challenging
12Shunt Supply Resource Allocation
- Finite number of pins, metal lines for power
- Need to allocate resources between main and shunt
supplies - For resistive losses
- If ensure that Vshunt only handles transients
- Resistive losses of main supply not heavily
affected
13Quiescent Output Current
- Went to push-pull topology to minimize quiescent
current - But many designs have significant Istatic
- To ensure output devices are off when Vreg is
quiet - Use non-linear switching to control the output
stage
14Comparator Feedback with Dead-Band
- Use comparators in feedback path to generate
full-swing drive signals - To avoid limit cycle
- Offset thresholds to create dead-band
15Output Stage Design
- Requirements on output stage
- Needs low td,on
- To maintain voltage- positioned response
- Needs good supply rejection
- Vshunt will be noisy
- Needs to burn minimal static power
- To meet efficiency goal
16Replica Biased Output Stage
- Replica loop sets gate bias (Vbias)
- Minimal current spent in feedback amplifier for
efficiency - High impedance amp add buffer between Vbias and
switching node
17Switched Source Follower Buffer
- Source follower Mp_sf used to isolate Vbias
- Turned on only when pushing current
- But, waiting for Isf to charge Vgn too slow
- Mp_up turned on during transition
18Outline
- Regulator Topology
- Regulator Design
- Experimental Verification
- Conclusions
19Test-Chip Details
- 65nm SOI AMD test-chip
- Regulator uses same power distribution scheme as
processor - With reallocation for Vshunt
- On-chip noise generator and perf. monitor for
testing
Processor
Regulator Circuits
Scan/Config.
20Measured Results
- Regulator reduces broadband noise by 30
- Total power dissipation actually reduced by up to
1.4
21Impedance with Faster Process
- Process was in development
- Low device ft
- Measurement matches simulation with slower devices
- Expect to reach 50 noise and 4 power
reduction in production process with higher ft
22Outline
- Regulator Topology
- Regulator Design
- Experimental Verification
- Conclusions
23Conclusions
- To be adopted, on-chip regulation must not
increase total power - Can in fact build power-neutral regulator
- Push-pull topology with second supply
- Switched source-follower output stage
- Measured 30 noise reduction with no power
penalty - In fact, reduced power by 1.4
24Acknowledgements
- This work was funded in part by C2S2
(www.c2s2.org) - AMD Sunnyvale Design Center