Chart the course of post- Moore's Law chip design

Whoa. Just back from the most recent Palo Alto Colloquia over at Lockheed Martin's Advanced Technology Center.

Dr. Kwabena Boahen, Stanford University, is an extraordinarily articulate man of humble demeanor, possessed of some of the most noble goals in this segment of industry. "We have two synergistic goals: We wish to understand how brains work; this will enable us to replace damaged neural tissue. And we want to build computers that work like brains; this will enable us to increase computational power a million fold."

Reliance upon abstraction as the primary means of managing circuit complexity is approaching the boundaries of its useful limits as, "Nanoelectronic technology promises to cram a trillion transistors onto a 1 cm² chip." Nature provides instructive alternatives for managing such complexity.

A couple of pivotal conceptual breakthroughs that caught my attention:

• Don't morph neural circuits into silicon, morph the rules that build the circuits.
• Once a chip leaves the fab, no circuit changes are possible. Softwires make post-production self-configuring circuits possible.
  

I might briefly and crudely describe softwires as RAM gates, but the slides do a much better job of illustrating this innovation. The slides are going up on internal Lockheed site in the next day or so, but public access won't be available until next week, so be sure to check back for them. This is mostly pre-hippocampal emulation in silicon, but the NeuroGrid project will significantly expand applied understanding and emulation of the brain's processes that create meaning and context from received stimuli.

At least, that's how I understood the talk.

too.late