A Throughput-Agnostic 11.9-13.6 GOPS/mW Multi-Signal Classification SoC for Cognitive Radios in 40nm CMOS

Speaker: Sina Basir-Kazeruni
Affiliation: Ph.D. Candidate

Department Research Forum – Circuits and Embedded Systems

Hosted by Prof. Danijela Cabric


Abstract: A blind classification SoC for cognitive radios, featuring multi-signal channelization, 16-core dynamic parallelism-frequency scaling and GALS-based multithreading, is realized in 40nm CMOS. Targeting ≥95% detection probability and ≤0.5% false-alarm rate, the SoC achieves a throughput-insensitive energy efficiency of 11.9-13.6GOPS/mW for multiple 7.8-125MHz bandwidth-agnostic signals in a 500MHz channel. The SoC shows 2.1× lower energy, >2.7× less efficiency variation, 1.2× baseband area and up to 4× processing time reduction compared to prior work. Throughput-matched scheduling of spatial resources enables operation at peak energy efficiency without the need for any voltage adjustment.

Date(s) - Oct 28, 2015
12:00 pm - 1:00 pm

E-IV Tesla Room #53-125
420 Westwood Plaza - 5th Flr., Los Angeles CA 90095