Future of Analog Design and Upcoming Challenges in Nanometer CMOS Greg Taylor VLSI Design 2010
Outline Introduction Logic processing trends Analog design trends Analog design challenge Approaches Conclusion 2
Introduction This talk will focus on analog design on digital processes Small geometries Low voltages Integration with logic Processes optimized for analog circuits are a different problem 3
Intel view: Introduction (cont) Don t get in the way of digital scaling Transistor count doubles every 24 months Don t violate the Law! Alternative view Logic processes are leading the way to smaller geometries They provide the first look at the design of analog circuits at these feature sizes 4
Outline Introduction Logic processing trends Analog design trends Analog design challenge Approaches Conclusion 5
Logic Processing Trends Trans sistor count 10 9 10 7 10 5 2x every 2 years 10 9 10 7 10 5 10 3 10 3 1970 1980 1990 2000 2010 2020 Performance and functionality continue to improve with increased transistor count 6
SRAM Scaling 10 Cell Are ea (um 2 ) 1 0.5x every 2 years 65 nm, 0.570 um 2 45 nm, 0.346 um 2 0.1 1995 2000 2005 2010 32 nm, 0.171 um 2 Mark Bohr: IDF 2009 22 nm, 0.092 um 2 7
Contact 1978 30 Years of Scaling Ten 32nm SRAM Cells 2008 1 µ m 8
Process trends: New Generation Every 2 Years 10 10000 Microns 1 365 0.7X every 2 years 130nm 248 90nm 193 65nm 45nm 32nm 22nm 0.1 Nanotechnology (< 100nm) 1000 100 Nanometers Nominal feature size 13 0.01 1970 10 1980 1990 2000 2010 2020 Source: Intel 9
Yield Trends Defect Density (log scale) 90nm 65nm 45nm 32nm 2 year 2002 2003 2004 2005 2006 2007 2008 2009 2010 Higher Chip Yield 2 year technology cycles High yields Fast ramp to volume Progress is not slowing 10
Outline Introduction Logic processing trends Analog design trends Analog design challenge Approaches Conclusion 11
Why Analog Design? (analog circuits are needed to interface with reality. reality is analog) Magical Fairyland of 1 s and 0 s Thermal Configuration Temporal Electrical Source: Rachael Parker 12
Growth in Analog Circuits Analog Circuits/Systems on Intel Microprocessors Microprocessor: clock generator IO bus thermal shutdown Circuits 35 30 trim 25 Multi-core SOC: multi-domain clocking 10-20 PLLs high speed serial IO low jitter clock # of Unique Analog 20 15 10 5 Fuse Thermal Clocking IO advanced thermal and power management unit-level trim of analog components 0 250 180 130 90 65 45 32 Process Node (nm) 13
Performance Requirements ~10x over a decade Pentium II Processor 250 nm PLL: 400 MHz Fmax DTS: 140 C ± 15 C single trip point I/O: 266/400 MT/s Core i7 Processor 45 nm PLL: 3+ GHz Fmax DTS: - 10 C to 140 C 1 C Resolution I/O: 6400 MT/s 14
Outline Introduction Logic processing trends Analog design trends Analog design challenge Approaches Conclusion 15
10 Voltage Scaling Voltage (V) 1 0.7x/gen 2000 1000 800 500 350 250 180 130 90 65 45 32 CD (nm) Voltage scaling has slowed on recent technologies This is the technology maximum voltage 16
Analog Scaling with Voltage Reduced operating range of classical circuits Signal shrinks Noise doesn t Low overdrive exacerbates Vt mismatch Weakly off switches leak 17
Device Mismatch Trend Transistor threshold variation increases with shrinking device size σv t = C 2 / W eff eff L eff Process improvements provide some relief Scaled device sizes still lead to variation increases 18
Scaling of σvt Random Variation Kuhn, Reducing Variation in Advanced Logic Technologies, IEDM 2007 19
Thermal noise 1/f noise dv dv 2 4kT g 2 eq = / m KF = WL C eq 2 ox 2 Noise 3 df df f Dynamic range decreases for smaller L and W Sansen, Analog IC Design in Nanometer CMOS Technologies, VLSI Design 2009 L W 20
Co-Optimization Time to market can force design in parallel with process development The good news is that process development is in parallel with design Reserve some adaptability to cover when development does not go as expected Note: Digital circuits generally have analog success criteria Analog circuits usually have binary success criteria 21
Other Challenges Mixed Signal Validation: When digital and analog circuits are mixed the validation approaches that are effective for either in isolation fail to adequately cover the combination Circuit simulation runs times explode RTL simulation doesn t model analog behavior Testability Increasingly difficult to characterize the clock Low bandwidth, legacy DFT pins Difficult to do volume analog test, yet statistical design requires statistical test 22
Outline Introduction Logic processing trends Analog design trends Analog design challenge Approaches Conclusion 23
Avoid Analog If You Can Many functions can be implemented with analog or digital approaches If possible, choose digital Cntr 24
Copy* It If You Can t Avoid It A new implementation is fun It s also a way to find new kinds of mistakes to make It s more expensive and takes longer than copying Take advantage of the work that others have done to find mistakes and validate solutions * Paying attention to IP laws 25
If You Can t Copy It, Then Apply Good Design Practices Keep It Simple! As simple as possible, as digital as possible Document your work The flip side of reuse, is that you need to make your own work reusable Documentation improves the quality of design reviews, helping find mistakes sooner People who are reusing your work are invested in finding errors 26
Good Design Practices If you re co-developing along with your process, build in lots of tolerance for process targeting Build in tolerance for variation There are only so many atoms available in those transistors Keep matching localized Utilize self calibration, trimming, and fuse options This will help increase tolerance to retargeting and variation 27
If You Can t Fix It, Feature It Analog to digital converter Vin + Daly, A 6b 0.2-to-0.9V Highly Digital Flash ADC with Comparator Redundancy, ISSCC 2008 Adaptive frequency clocking time Kurd, Next Generation Intel Micro-architecture (Nehalem) Clocking Architecture, JSSC 2009 Frequency voltage freq Voltage 28
Statistical Design Increasing device variation requires variation aware design Worst case design is generally not practical Skew corner simulation does not highlight the impact of within die variation Statistical design techniques help predict and understand the impact of variation Monte Carlo Design of Experiments These tools don t replace the need for engineers to understand statistics! 29
Validation Is Essential Circuit simulation It s the time honored approach Necessary, but not sufficient in a mixed signal system Mixed signal validation RTL is often discrete time, discrete voltage Hybrid circuits control RTL invisible behavior Impedance, delay, temperature, voltage, current Design reviews Reviewer team needs to have variety and engagement 30
AMS Validation Ensure that individual analog blocks work The traditional realm of circuit simulation Ensure that analog blocks work together Not just at the center of the spec range Ensure that digital and analog work together Both control and data flow 31
Other Rules of Thumb Need to enforce supply isolation between analog and digital circuits Production vs. simulation schematics It s tempting to make alternate schematics for simulation that include extras Don t If necessary build a test bed in a higher level of hierarchy 32
Outline Introduction Logic processing trends Analog design trends Analog design challenge Approaches Conclusion 33
Conclusion We are increasing the scope and complexity of analog circuits on logic processes At the same time those processes are becoming harder to work with Eliminate unnecessary analog design Avoid making analog the limiter where possible floorplan constraints, timing margin Mitigate process scaling non-idealities Trim, offset cancellation, noise shaping, high voltage analog, etc Don t skimp on mixed signal validation 34
Questions?