Scientists use nanotechnology to try building computers modeled after the brain
(Nanowerk Spotlight) Scientists have great expectations that nanotechnologies will bring them closer to the goal of creating computer systems that can simulate and emulate the brain's abilities for sensation, perception, action, interaction and cognition while rivaling its low power consumption and compact size.
DARPA for instance, the U.S. military's research outfit known for projects that are pushing the envelope on what is technologically possible, has a program called SyNAPSE that is trying to develop electronic neuromorphic machine technology that scales to biological levels. Started in late 2008 and funded with $4.9 million, the goal of the initial phase of the SyNAPSE project is to "develop nanometer scale electronic synaptic components capable of adapting the connection strength between two neurons in a manner analogous to that seen in biological systems, as well as, simulate the utility of these synaptic components in core microcircuits that support the overall system architecture."
Independent from this military-inspired research, nanotechnology researchers in France have developed a hybrid nanoparticle-organic transistor that can mimic the main functionalities of a synapse. This organic transistor, based on pentacene and gold nanoparticles and termed NOMFET (Nanoparticle Organic Memory Field-Effect Transistor), has opened the way to new generations of neuro-inspired computers, capable of responding in a manner similar to the nervous system.
"Basically, we have demonstrated that electric charges flowing through a mixture of an organic semiconductor and metallic nanoparticles can behave the same way as neurotransmitters through a synaptic connection in the brain," Dominique Vuillaume, a research director at CNRS and head of the Molecular Nanostructures & Devices group at the Institute for Electronics Microelectronics and Nanotechnology (IEMN) tells Nanowerk.
In the development of tomorrow's computing systems, one approach consists in mimicking the way biological systems such as neuron networks operate to produce electronic circuits with new features. In the nervous system, a synapse is the junction between two neurons, enabling the transmission of electric messages from one neuron to another and the adaptation of the message as a function of the nature of the incoming signal.
This figure illustrates the comparison of a synapse with the NOMFET. (Image: Dr. Vuillaume, IEMN-CNRS)
Vuillaume explains the underlying biological working principle of their transistor: In a spiking biological synapse, the synapse transforms a spike of action potential arriving from the presynaptic neuron into a chemical discharge of neurotransmitters detected by the post-synaptic neuron and transformed into a new spike. Each spike at the synapse input activates a fraction of these neurotransmitters and the amplitude of the transmitted spike is a function of this fraction. Then this fraction of neurotransmitters is inhibited – and not available to transmit a second spike – and is then recovered with a characteristic time constant (typically in the range of seconds).
As a consequence, the response of a synapse to a train of pulses depends on the time interval between successive pulses that determines the amount of available neurotransmitters. Depending on the nature of the synapse, the response to a constant frequency train of pulses can be either depressing (amplitude of the successive output spikes decreasing) or facilitating (amplitude increasing).
"This is a fundamental property of the synapse behavior called STP (short-term plasticity)" explains Vuillaume. "In our NOMFET, the presynaptic signal is the pulse voltage applied on the NOMFET and the output signal is the drain current. The holes – the charge carriers in the p-type organic semiconductor used here – trapped in the metallic nanoparticles play the role of the inhibited neurotransmitters. The number of trapped holes in the nanoparticles depends on the input spike voltage and the output signal, the current, is a decreasing function of the number of holes stored in these nanoparticles – because less holes are available to transmit the current through the NOMFET."
At each spike, a certain amount of holes are trapped in the nanoparticles. Between pulses the system relaxes: the holes escape with a characteristic time that we have adjusted to few seconds (same as in the biological synapse) by optimizing the nanoparticles and device geometries. Thus, depending on the frequency of the input voltage spikes, the output of the NOMFET is able to reproduce the depressing or facilitating behavior of the synapse.
For example, if the synapse receives very closely packed pulses of incoming signals, it will transmit a more intense action potential. Conversely, if the pulses are spaced farther apart, the action potential will be weaker. It is this plasticity that the researchers have succeeding in mimicking with the NOMFET.
In other words, the encapsulated gold nanoparticles, fixed in the channel of the transistor and coated with pentacene, have a memory effect that allows them to mimic the way a synapse works during the transmission of action potentials between two neurons. This property therefore makes the electronic component capable of evolving as a function of the system in which it is placed.
According to Vuillaume, this appears to be the first demonstration that a single electronic device can mimic the STP property of the biological spiking synapse.
"In our group" says Vuillaume, "the main motivations are the design and characterization of molecular and nanoscale electronic devices, the elucidation of the fundamental electronic properties of these molecular and nanoscale devices, the study of functional molecular devices and integrated molecular systems and the exploration of new computing paradigms using molecules and nanostructures. This work relates to this last point and more specifically, the main motivation was to build nanoscale devices that can be used in neuron-inspired computers."
He notes that the human brain contain more synapses than neurons (by a factor of about 10,000), and therefore it is necessary to develop a nanoscale, low power, synapse-like device if scientists want to scale neuromorphic circuits towards the human brain level.
"This feature has recently prompted the research for nanoscale synaptic devices" says Vuillaume. "In fact, neural networks have been already developed and used in some applications. However, even if silicon CMOS chips have been designed and fabricated to emulate the brain behaviors, this approach is limited because it takes several – at least seven – silicon transistors to build an electronic synapse. Here, we did the same job with a single device."
The potential application of this work is to increase the performances of neural-network computing circuits. Moreover, nanoparticles and molecules are nanosize objects suitable for nanodevice fabrication; they can be manipulated and assembled by low-cost, bottom-up, techniques (e.g., self-assembly); and they are prone to work on flexible, plastic, substrates.
Vuillaume points out that this later feature might be advantageous if we envision connecting artificial neuromorphic devices and circuits (based on the NOMEFT) with soft biological materials. However, he also notes that, even if this work reduces the number of electronic devices required to mimic a biological synapse, the high level of synapse-neuron connectivity in the brain requires the development of artificial neural networks computing circuits in 3D.
"We believe that this challenge is probably more attainable using molecules, nanoparticles, self-assembly and bottom-up approaches than by conventional silicon nanotechnologies" Vuillaume concludes.