Event-Driven QDI Circuits Nabil Imam 1 , Filipp Akopyan 2 , John - - PowerPoint PPT Presentation

▶

Apr 03, 2024 339 likes •513 views

A Digital Neurosynaptic Core Using Event-Driven QDI Circuits Nabil Imam 1 , Filipp Akopyan 2 , John Arthur 2 , Paul Merolla 2 , Rajit Manohar 1 , Dharmendra Modha 2 1 Cornell University 2 IBM Research Almaden Neuromorphic VLSI Biological nerve

SLIDE 1

A Digital Neurosynaptic Core Using Event-Driven QDI Circuits

Nabil Imam1, Filipp Akopyan2, John Arthur2, Paul Merolla2, Rajit Manohar1, Dharmendra Modha2

1Cornell University 2IBM Research Almaden

SLIDE 2

Neuromorphic VLSI

05/7/2012 – 2/16

Biological nerve cell models in silicon
Real-time, low-power operation
Commercial and medical applications
Tools for brain science

SLIDE 3

Spiking Neural Networks

TIME

05/7/2012 – 3/16

Neuron
Dendrites (inputs)
Axon (output)
Synapses (connections)

SLIDE 4

Models in Silicon

Differential Equations (e.g. Hodgkin-Huxley, Integrate-and-Fire, etc) Analog Circuits

Computatation
Intrinsic properties
Network connectivity
Efficiency
~1009 neurons and ~10012 synapses in the human brain
20 W Power
2 L Volume

05/7/2012 – 4/16

SLIDE 5

Computation Memory

Off-chip Memory For Synaptic Information

Implementation Challenges

05/7/2012 – 5/16

Communication

AER Packet

SLIDE 6

Crossbar Synapses

05/7/2012 – 6/16

Integrated Computation

and Memory

Crossbar

– Rows: Axons – Columns: Dendrites – Junctions: Synapses

Flexibility
Efficiency: Large fanout

in one operation

SLIDE 7

Scalable Neurosynaptic Core

Crossbar Synapses
Digital Circuits

– Deterministic – Scalable (45nm SOI)

Asynchronous design

– Minimal power dissipation when idle – Fast when active – Robust

Scalable to a multi-core/multi-chip system

05/7/2012 – 7/16

SLIDE 8

System Architecture

04/11/2012 – 8/14

SLIDE 9

Neuron Model

05/7/2012 – 9/16

฀ if Vi(t)   Spike 1 Vi(t 1)  0

฀ V(t 1) V(t)  Leak  [A j(t) W j  S j

j1 k



]

Leaky Integrate-And-Fire

SLIDE 10

Communication Infrastructure

AER transmitter and receiver for fan-out Routing fabric for multi-core communication Scheduler for synchronization

05/7/2012 – 10/16

SLIDE 11

Chip Test

05/7/2012 – 11/16

SLIDE 12

Example Application:

Sound Localization

05/7/2012 – 12/16

SLIDE 13

Example Application:

Odor Recognition

Neuron Index [C]

05/7/2012 – 13/16

SLIDE 14

Example Applications

Pong! Motor Control Auto-associative Memory Visual Recognition

05/7/2012 – 14/16

SLIDE 15

Conclusion

Neurosynaptic Core

– Digital Asynchronous Circuits – Crossbar synapses – Scalable to multiple cores and multiple chips

Applications

– 1-1 hardware-software correspondence – Sensory-motor activity – Bio-inspired solutions to technological problems

05/7/2012 – 15/16

SLIDE 16

A Digital Neurosynaptic Core Using Event-Driven QDI Circuits

Nabil Imam1, Filipp Akopyan2, John Arthur2, Paul Merolla2, Rajit Manohar1, Dharmendra Modha2

1Cornell University 2IBM Research Almaden

05/7/2012 – 16/16