Condensed Matter Physics Seminar

  • CNST/UMD Postdoctoral Researcher
  • Electron Physics Group
  • National Institute of Standards and Technology (NIST)

Using the Dynamics of Nanodevices for Artificial Intelligence

Artificial  neural  networks  are  performing  tasks,  such  as  image  recognition  and  natural language  processing,  that  offer  great  promises  for  artificial  intelligence.  However,  these algorithms run on traditional computers and consume orders of magnitude more energy more than the brain does at the same task. One promising path to reduce the energy consumption is to build dedicated hardware to perform artificial intelligence. Nanodevices are particularly 
interesting because they allow for complex functionality with low energy consumption and small size.

I discuss two nanodevices. First, I focus on stochastic magnetic tunnel junctions, which can emulate the spike trains emitted by neurons with a switching rate that can be controlled  by  an  input.  Networks  of  these  tunnel  junctions  can  be  combined  with  CMOS circuitry to implement population coding to build low power computing systems capable of processing  sensory  input  and  controlling  output  behavior.  Second,  I  turn  to  different nanodevices, memristors, to implement a different type of computation occurring in nature: swarm intelligence.  A broad class of algorithms inspired by the behavior of swarms have been proven successful at solving optimization problems (for example an ant colony can solve a maze). Networks of memristors can perform swarm intelligence and find the shortest paths in mazes,  without  any  supervision  or  training.  These  results  are  striking  illustrations  of  how matching the functionalities of nanodevices with relevant properties of natural systems open the way to low power hardware implementations of difficult computing problems.

For More Information, Please Contact: