Nanoscale Memristor Mimics Biological Synapse

The existence of the memristor, which was first proposed in 1971, was proved through groundbreaking research performed by researchers at HP Labs. The memristor is the fourth fundamental element of electronic circuits apart from resistors, capacitors, and inductors and as the name memory register suggests, it can retain information even after the power is turned off. Memristors can adopt high- or low-resistance states, which are stable and this key feature allows them to act as nonvolatile memory. It is expected that memristors could one day be used to develop computer systems that have memories similar to biological systems.

Research work conducted by Wei Lu, assistant professor of electrical engineering and computer science at the University of Michigan, Ann Arbor, and his team has provoked renewed interest among the research community in the ability of memristor devices to emulate neural learning. They have discovered that the memristors can behave just like synapses. For instance, they respond to neuron spikes and store information such as biological synapses and can connect a large number of neurons together like biological synapses. These findings by Lu’s team show that it is now possible to build a brain-like computer using electronic components, namely, transistors and memristors.

The key was to identify the similarity between synapses and memristors. In a mammalian brain the computing units, neurons, are connected to each other through programmable junctions called synapses. The synaptic weight modulates how signals are transmitted between neurons and can in turn be precisely adjusted by the ionic flow through the synapse. A memristor by definition is a resistive device with inherent memory. It is in fact very similar to a synapse--they are both two-terminal devices whose conductance can be modulated by external stimuli with the ability to store (memorize) the new information.

In their study, a nanoscale silicon-based memristor was fabricated to mimic the synapse. The silicon memristor consists of a pair of electrodes that sandwich an amorphous-silicon layer doped with silver (Ag) atoms, with high-Ag concentration near the top electrode and low-Ag concentration near the bottom electrode. When a positive voltage is applied across the memristor, the Ag ions in the silicon layer will drift to the bottom electrode and increase the overall conductance of the device, and vice versa. The new conductance state is maintained until the next voltage pulse is applied. By controlling the Ag doping profile and other device parameters, the team was able to show that the change in the memristor conductance is proportional to the time integral of the voltage applied across it. In other words, the device state is not determined by the existing signals, but by the history of the applied signals. This device behavior is consistent with the flux-controlled memristor device model first proposed by Leon Chua in 1971.

Furthermore, this property enabled the researchers to precisely control the memristor conductance with external stimuli. The longer the voltage pulse is applied across the memristor, larger is the change in conductance observed. These properties essentially enable the memristor to mimic synaptic action. For instance, in their paper that was published in the journal Nano Letters in March 2010, the team has demonstrated that a hybrid electrical circuit consisting of complementary metal oxide semiconductor (CMOS) "neurons" and memristor synapses can achieve spike-timing dependent plasticity (STDP), an important synaptic function.

Besides their potential use in neuromorphic circuits to build brain-like computers, memristors such as the ones Lu’s team has developed could help in building faster and better circuits. Primarily, they can provide high-density storage needed for memory applications in conventional circuits. Secondly, new approaches to build circuits may be developed so that the increase in computing power does not come from the increase in raw device speed (clock frequency), but comes from the increase in computing efficiency instead.

As the next step, the team intends to build larger circuits that consist of hundreds of CMOS neurons and memristor synapses. There are a number of issues that still need to be addressed. These include device variation, heat dissipation, and potential interference between devices. There has been rapid progress in memristor research to show that these obstacles may be overcome with proper understanding of the device operation and the circuit design.

Details:

Dr. Wei Lu

Assistant Professor

Electrical Engineering and Computer Science

University of Michigan

2242 EECS Building

1301 Beal Avenue

Ann Arbor, MI 48109

Phone: 734-615-2306

E-mail: wluee@eecs.umich.edu

To comment on this article, write to us at tiresearch@frost.com

To find out more about Technical Insights and our Alerts, subscriptions and research services, access http://ti.frost.com