IEEE Solid-States Circuits Magazine - Spring 2019 - 130

SSCS Singapore Chapter Hosts Technical Seminar
by Prof. Boris Murmann

The IEEE Solid-State Circuits Society
(SSCS) Singapore Chapter organized
the technical seminar, "Mixed-Signal
Circuit Design for the Data-Driven
World," presented by Prof. Boris Murmann, Stanford University, California.
The lecture was held at the Nanyang
Technological University, Singapore,
on 13 December 2018. Eighty-four
participants from industry, research,
and academia attended.
The lecture began with an overview of research topics from the
Murmann Mixed-Signal Group at Stanford University. Murmann explained
the motivation behind his recent research on mixed-signal interface
designs with innovations driven by
application-level and system-specific
customizations.
Next, he discussed some of his
recent research projects, including
data-compressive neural interfaces,
a sub-Nyquist digital predistortion
receiver, a data-compressive imager
for object detection, and mixed-signal
convolutional neural network (NN)
accelerators. Murmann talked about
novel neural interfaces utilizing
the sparsity of neural spike signals.
The efficient neural interface with
a massive number of read channels
(~10,000) was designed to achieve

Digital Object Identifier 10.1109/MSSC.2019.2909375
Date of publication: 24 June 2019

Prof. Murmann presents his talk "Mixed-Signal Circuit Design for the Data-Driven World,"
hosted by the SSCS Singapore Chapter.

greater than 2 Gb/s of the aggregate
data rate with 10 mW of the limited
power budget. He also discussed the
sub-Nyquist sampling-based digital
predistortion receiver, which was
developed to resolve the growing
bandwidth requirement of a power
amplifier digital-predistortion sensing path in modern wideband radiofrequency systems.
In addition, Murmann presented
a histogram of an oriented gradientbased object-detection system with
logarithmic gradients, which can
reduce the power and complexity of
the interface between the imager and
the off-chip digital processor. Murmann further introduced the mixedsignal-based binary convolutional
NN (CNN) processor, built with an eye

Prof. Murmann (center left) with lecture attendees.

130

S P R I N G 2 0 19

IEEE SOLID-STATE CIRCUITS MAGAZINE

toward recent algorithmic advancements and targeting low-complexity
applications. With architectural parallelism, parameter reuse, and the use
of energy-efficient mixed-signal processing for data accumulation, the
CNN processor has achieved high
accuracy (86%) on a Canadian Institute for Advanced Research-10 data
set, while occupying 328 kB of onchip memory and consuming 3.8-µJ
energy per classification.
After explaining the four specific
research examples, Murmann summarized his talk with some insightful comments. The technical seminar was followed by an interactive
Q&A session.
-Bongjin Kim
SSCS Singapore Chapter

IEEE Solid-States Circuits Magazine - Spring 2019

Table of Contents for the Digital Edition of IEEE Solid-States Circuits Magazine - Spring 2019