Information Theoretical Approach to Analog and Mixed-Signal IC Design
Zeljko Ignjatovic, Professor, Electrical and Computer Engineering Department, University of Rochester
Wednesday, October 7, 2009
As CMOS technology continues to scale benefiting from the enormous momentum of the digital IC market, analog and mixed-signal circuit designers are facing more difficulties. The technology is optimized for digital designs in primarily one tradeoff between circuit speed and power consumption. In contrast, analog and mixed-signal circuits require a multidimensional design space where almost all two circuit parameters trade with each other. Short-channel effects lead to greater shifts in thresholds voltage, increased noise and mismatch. The velocity saturation limits the mobility and obtainable transconductance and high-speed performance of analog amplifiers. Lower power supply voltage limits the dynamic range. In the realm of digital CMOS technologies, the conventional approaches to analog IC design and optimization may no longer be sufficient to guarantee required performance. In our work, we propose an alternative approach in designing an analog IC, where the circuit is described as a communication/information channel. Similar to a communication system, an IC is supplied with an input analog signal by an information source. It has a means to perform source and channel coding via sampling, filtering, and modulation circuitry. It injects noise such as electronic circuit noise, quantization noise, non-linear distortion, power supply and substrate noise, which limits the amount of information that can be transmitted through the IC. Also, it has a means to decode the output from the channel (such as the digital decimation filter in Sigma-delta ADCs). Finally, it provides decoded information to its user. Once the analog IC is described as communication channel, we are able to define its capacity (upper performance limits) and find optimal design solutions to reach the capacity. Our research represents a synthesis of principles from previously disconnected research areas: Information Theory, Telecommunications, Signal Processing, Detection and Estimation Theory on one side and Analog and Mixed-signal IC design on the other. This novel approach allows us to design more reliable analog and mixed-signal IC that are less susceptible to technology scaling. The information theoretical approach will be demonstrated through several design examples including accelerometer readout with less than 20uW of power, fully digital image sensors with focal-plane compression and readout noise levels that are significantly lower than any of the previously reported designs, Sigma-delta A/D converters insensitive to power supply and substrate coupling noise, and our most recent work on Turbo-code based A/D converters.
Zeljko Ignjatovic received the B.Sc. degree in electrical engineering from University of Novi Sad, Serbia, and the M.Sc. and Ph.D degrees in electrical engineering from University of Rochester, NY in 1999, 2001, and 2004, respectively. He is currently an Assistant Professor of Electrical and Computer Engineering at University of Rochester. His areas of interest include analog circuit design and development of A/D converters, image sensors and related signal processing methods.