|
Jennifer Hasler
|
Terry Stewart
| |
Georgia Institute of Technology
|
University of Waterloo
|
This topic integrates analog circuit design and analog computing
and high-level tools to enable many designers
towards building neuromorphic engineering systems.
| |
Our goal is to build an analog neuromorphic hardware community fully
immersed in developing the tools and infrastructure needed to develop
mature analog neuromorphic systems.
|
|
Summary:
This topic area
builds an analog neuromorphic hardware community
fully immersed in developing
the tools and infrastructure needed to develop mature analog neuromorphic systems.
High-level tools
enable
low-level principles of analog design
towards neuromorphic system design.
New tool development,
as well as new hardware development,
supports large-scale analog iand neuromorphic systems.
Tools are essential for a wide user base to use neuromorphic hardware
that is often programmable
(e.g. FG devices)
as well as configurable
(e.g. large-scale Field Programmable Analog Arrays (FPAA) )
for their particular application.
|
|
Neuromorphic Hardware / Software Platforms:
- FPAA Boards with remote interfacing:
The USB-interfaced FPAA boards will be utilized as a general framework
to create individual circuit designs,
as well as getting a vision of the capability
and realistic viewpoint of working with current neuromorphic hardware.
- Braindrop boards (remote?)
- Nengo and Nengo-Braindrop
|
FG = Floating-Gate, as in FG Devices, FG Circuits, FG Systems
FPAA = large-scale Field Programmable Analog Arrays, as in SoC FPAA
Workshop Summary :
We will have multiple sessions to teach hardware interfacing (e.g. FPAA devices),
as well as teaching sessions to understand core tool concepts,
as well as fundamental circuit concepts.
- July 28: Overview of ANT Track: Terry Stewart
- July 1: Overview Lecture:
Jennifer Hasler,
" Neuromorphic Hardware and Tools"
- July 1, Discussion Session
- July 2, Invited lecture (Kwabena Boahen):
- more sessions TBD
Discussion
sessions
during Virtual Telluride three weeks
|
|
Track Lectures:
- Underlying theory of analog design through the neuromorphic design history.
- Tools enabling design applied towards the design, synthesis, and verification
of neuromorphic systems
- Interactive (mostly virtual) sessions
enabling
hand-on work with existing analog
systems, via
remote access to FPAAs and Braindrop,
low-level access and high-level programming (e.g. Nengo to program Braindrop),
as well as access to simulating these analog systems.
|
Potential Projects
Projects:
apply analog circuit and systems to solve neuromorphic application problems.
Some projects
emphasize on function approximation / classification,
such as command-word / keyword spotting.
Other projects further develop the existing tools to better
support these particular applications.
|
- Text classification using FPAA / Braindrop / SPICE simulation
- Characterizing the space of functions that can be well-approximated with different hardware (and how adjusting that hardware would affect that space of functions)
- Expanding an analog standard cell library
- Developing higher level abstraction tools for neuromorphic hardware programming
(e.g. FPAA, Braindrop)
|
Videos and Reading on Hardware, Tools, and Neuromorphic Design
Neuromorphic Overview and Roadmap
IEEE
Spectrum (2017),
Neuromorphic Comp (2013).
FPAA Devices, Systems, and Tools
Short videos:
-
FPAA
History, Dev, Classification, and Directions
-
Historical
perspective on FG for Neuro Eng:
- FPAA Enabling Physical
Computing
- Open-Source FPAA
tools
- Potential FPAA Capabilities:
- Starting point on analog standard-cell
libraries
FPAA Workshop
material
FPAA Tool Download :
Ubuntu VM
Core Reading Material
-
FPAA Overview
(IEEE Proceedings 2020).
-
Open-source FPAA tool
overview
(WOSET 2020).
-
First System on Chip (SoC) FPAA
IC
(TVLSI 2016).
- On-chip VMM+WTA learning
classifier
(ECAS 2018).
- FPAA Remote
System
(JPLEA 2016)
- Additional
resources
|
Introduction to Analog IC Design
Neuromorphic Design using Nengo
|
|