MSc thesis project proposal

[2018] Subthreshold Stochastic Computing for On-Chip Low Energy BioSignal Processing

Stochastic computing is a digital computing paradigm in which the values in the system are represented by streams of random bits. In such a system some complex computations may be realized by simple bit-wise operations.

Low energy operation is one of the most important metrics for the implantable neural recording microchips. As the energy supplied to such a chip through electromagnetic waves, ultrasound, energy harvesting or batteries is very limited, using the available energy in the most efficient way is of paramount importance.

In the past, it has been shown that stochastic computing was not energy efficient enough when compared to standard CMOS digital implementation. In this project, we would like to combine subthreshold and possibly asynchronous operation of CMOS digital gates with stochastic computing and biosignal (EEG, ECG, ECOG, etc.) specific optimizations. As signal specific methods such as sampling frequency or precision reduction may be employed in realtime, there is a possibility that stochastic computing combined with subthreshold (and asynchronous) operation may be more energy efficient.

Assignment

The exact timeline of the project will be determined in close collaboration with the student. During this project, first, stochastic computing will be studied and circuit macros based on the recent research will be implemented. A previously developed high-level subthreshold energy model will be used for both the energy/performance evaluation and signal-specific optimization of stochastic computing circuitry. Based on the evaluations and simulations, new stochastic computing circuitry will be developed and implemented in a standard CMOS technology, whose feature size will be decided during the course of the project. Two of the crucial blocks during the design of the stochastic computing system will be the random number and stochastic number generators and special emphasis will be on these blocks. There is also a possibility that this project will be combined with the output of the ongoing project in the BioElectronics Group about stochastic resonance.

Requirements

For this project we are looking for a student with a strong mathematical background with previous experience with both mixed-signal design and digital design flow. A good understanding of device physics and noise characteristics is recommended.

Experience with biosignals and a scripting language such as Python is a plus.

Contact

dr. Can Akgun

Bioelectronics Group

Department of Microelectronics

Last modified: 2018-03-14