News Silver Nanowire Networks to Overdrive AI Acceleration, Reservoir Computing

[Admin]

The market thirsts for ever-increasing AI acceleration and training capabilities, and it seems safe to assume that Nvidia's offerings - and its generational performance improvements - aren't enough for the future of AI that we're seemingly heading towards - one throughout (HPC), cloud, personal computing and edge computing. Like fully analog computing, Silver Nanowires and nanowire networks in general are neuromorphic alternatives to our current AI paradigm.

Silver Nanowire Networks to Overdrive AI Acceleration, Reservoir Computing : Read more

 

[bit_user]

Wow, this sounds pretty impressive! I wonder how temperature-sensitive it is.

If there’s one area where biological processing units still are miles ahead of their artificial (synthetic) counterparts, it is energy efficiency.

I thought that had a lot to do with analog vs. digital and the fact that biology uses spiking neural networks, rather than the systolic approach currently dominating the AI industry.

 

[P.Amini]

This is interesting, we started in analog domain then we went digital and finally it surpassed analog in almost every area. Then that digital technology advancements led to AI, then advancements in AI is leading us to go back to analog again! I never would have thought of that!

 

[bit_user]

This is interesting, we started in analog domain then we went digital and finally it surpassed analog in almost every area. Then that digital technology advancements led to AI, then advancements in AI is leading us to go back to analog again! I never would have thought of that!

Throughout the history of Neuromorphic Computing, analog has always been lurking in the background.

Back in the early/mid-90's, Intel partnered with another company to build an analog IC for accelerating "multi-layer perceptrons". Wikipedia says the Ni1000 is the second generation chip - and a digital one - but references a previous generation that was analog:

​

Ni1000 - Wikipedia

en.wikipedia.org

​

Also, a family member once told me about a 512-synapse analog perceptron he worked on, in the research division of the USAF, back in the early 1960s.

I remember when I first learned about analog computers and how much faster they were than digital computers of the time. However, you have to be aware of their drawbacks and limitations. There are good reasons why the computing revolution very quickly went digital.

 

[USAFRet]

BINGO!!!