Contributing Editor, MEMS Investor Journal
Memristors use the molecular-level migration of dopants to change their conductance in response to current flow, thereby both measuring and remembering it. Like MEMSIC's accelerometer whose moving parts are molecules of hot air, memistors likewise use the motion of molecules to change their value. Does that make them a MEMS device? It's a matter of definition, but if MEMSIC's device is MEMS, then we may have to categorize memristors as MEMS too. Notably, the word memristor is not related to MEMS and is actually a portmanteau of "memory resistor".
Under the Defense Advance Research Projects Agency (DARPA) program called Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE), memristors could become the active element (artificial synapse) in brain-like layered architectures that processes information as it flows through them -- like the visual cortex.
In 2010, look for a flurry of announcements from DARPA's three SyNAPSE teams headed by IBM, Hewlett Packard (HP) and HRL (formerly Howard Hughes Labs). Recently, an HRL subcontractor, University of Michigan professor Wei Lu, published a paper describing his silver-based memristor which uses a standard complementary metal-oxide semiconductor (CMOS) process to craft artificial neurons connected by memristors as their synapses.
"We have shown that positive potentiating voltage pulses incrementally increase our memristors conductance, and that negative depressing voltage pulses incrementally decrease its conductance," said Lu. "We performed such conductance modulation up to 150 million times before our memristor showed significant degradation, which at a 1-Hz update rate would be about five years of continuous usage."
Lu's memristors are composed of silicon with silver atoms used as the dopant. By co-sputtering silver and silicon onto a memristive layer, current flowing through the device causes migration of the silver atoms into silver-rich (high conductivity) regions and a silver-poor (low conductivity) regions.
"These University of Michigan researchers are achieving similar results to our findings -- showing that memristors can very accurately emulate the spike timing dependent plasticity in real synapses," said HP Senior Fellow and director of its Information and Quantum Systems Lab.
HP already announced in 2008 that it can similarly change the conductance of its memristive material, and later this year will demonstrate its memristive material in various brain-like layered architectures. HP says that they have tried Lu's silver-based memristor material, but believe their material is longer lived. HP uses oxygen vacancies as the dopant in its titanium oxide memristor material.
Copyright 2010 MEMS Investor Journal