Neuromorphic function learning with carbon nanotube based synapses

Abstract

The principle of using nanoscale memory devices as artificial synapses in neuromorphic circuits is recognized as a promising way to build ground-breaking circuit architectures tolerant to defects and variability. Yet, actual experimental demonstrations of the neural network type of circuits based on non-conventional/non-CMOS memory devices and displaying function learning capabilities remain very scarce. We show here that carbon-nanotube-based memory elements can be used as artificial synapses, combined with conventional neurons and trained to perform functions through the application of a supervised learning algorithm. The same ensemble of eight devices can notably be trained multiple times to code successively any three-input linearly separable Boolean logic function despite device-to-device variability. This work thus represents one of the very few demonstrations of actual function learning with synapses based on nanoscale building blocks. The potential of such an approach for the parallel learning of multiple and more complex functions is also evaluated.

Nanoarchitectonic atomic switch networks for unconventional computing

“…However, further post-processing such as setting a threshold on the voltage to binarize the data can be performed to expand the device's response to a square wave input, a necessity for reliable Boolean logic computing. 38) It was found that the ASN was capable of replicating computing performances typical of reservoirs with 10 3 output signals. 39) Theoretical studies predicts the performance to scale with an increasing number of output signals due to the dependence on the regression algorithm.…”

Uncovering Cortical Modularity by Nanotechnology

“…MEAs using MWCNTs have the advantage of using a small size microelectrode with increased impedance and decreased charge-transfer capability (Gacem et al 2013). To decrease impedance, the effective surface area for recording of the electrode needs to be increased.…”

“…Furthermore, "converging technologies" that exploit the synergies between computer science, engineering, neuroscience, and psychology are envisioned to completely change the understanding of the entire science. Artificial synapses in neuromorphic circuits based on nanoscale memory devices have been recently accepted as a promising route for creating novel circuit architectures that tolerate variability and/or defects (Gacem et al 2013). Still, the implementations of neural network-type circuits that are based on non-CMOS (complementary metal-oxide-semiconductor) memory devices with learning capabilities are rare.…”

“…Still, the implementations of neural network-type circuits that are based on non-CMOS (complementary metal-oxide-semiconductor) memory devices with learning capabilities are rare. Gacem et al (2013) showed that memory elements based on CNTs may be used as "artificial synapses," combined with "conventional neurons" to be further trained to perform several functions (by applying a supervised learning algorithm). This is possible because the same device ensemble can be trained many times to code successively any type of 3-input combination of Boolean logic functions despite the variability among devices.…”

Ultrahigh Density Memristor Neural Crossbar for On-Chip Supervised Learning

“…However, since STDP is an unsupervised learning rule and is unsuitable for the learning of digital logic functions, these prototypes are not easily compatible with traditional CMOS digital chips. In contrast, neural-inspired logic blocks (NLBs) with supervised learning capability [5], [12]- [13] provide the programmability required for high density applications and attract an intensive research effort. Overall, the performance of existing neural networks can be evaluated in terms of the following four aspects [14]:…”