Anand Subramaniam

Hebbian Learning in Spiking Neural Networks With Nanocrystalline Silicon TFTs and Memristive Synapses

Characteristics similar to biological neurons are demonstrated in SPICE simulations of spiking neuron circuits comprised of submicron nanocrystalline silicon (nc-Si) thin-film transistors (TFTs). Utilizing these neuron circuits and corresponding device models, the properties of a two-neuron network are explored. The synaptic connection consists of a single nc-Si TFT and a memristor whose conductance determines the synaptic weight. During correlated spiking of the pre- and postsynaptic neurons, the strength of the synaptic connection increases. Conversely, it is diminished when the spiking is uncorrelated. This synaptic plasticity and Hebbian learning are essential for performing useful computation and adaptation in large-scale artificial neural networks. The importance of the result is augmented by the fact that these properties are demonstrated using models based on measured data from devices with potential for 3-D integration into a nanoscale architecture with extremely high device density.

Neural Learning Circuits Utilizing Nano-Crystalline Silicon Transistors and Memristors

Properties of neural circuits are demonstrated via SPICE simulations and their applications are discussed. The neuron and synapse subcircuits include ambipolar nano-crystalline silicon transistor and memristor device models based on measured data. Neuron circuit characteristics and the Hebbian synaptic learning rule are shown to be similar to biology. Changes in the average firing rate learning rule depending on various circuit parameters are also presented. The subcircuits are then connected into larger neural networks that demonstrate fundamental properties including associative learning and pulse coincidence detection. Learned extraction of a fundamental frequency component from noisy inputs is demonstrated. It is then shown that if the fundamental sinusoid of one neuron input is out of phase with the rest, its synaptic connection changes differently than the others. Such behavior indicates that the system can learn to detect which signals are important in the general population, and that there is a spike-timing-dependent component of the learning mechanism. Finally, future circuit design and considerations are discussed, including requirements for the memristive device.

Spike-Timing-Dependent Plasticity Using Biologically Realistic Action Potentials and Low-Temperature Materials

Spike-timing-dependent plasticity (STDP) is a fundamental learning rule observed in biological synapses that is desirable to replicate in neuromorphic electronic systems. Nanocrystalline-silicon thin film transistors (TFTs) and memristors can be fabricated at low temperatures, and are suitable for use in such systems because of their potential for high density, 3-D integration. In this paper, a compact and robust learning circuit that implements STDP using biologically realistic nonmodulated rectangular voltage pulses is demonstrated. This is accomplished through the use of a novel nanoparticle memory-TFT with short retention time at the output of the neuron circuit that drives memristive synapses. Similarities to biological measurements are examined with single and repeating spike pairs or different timing intervals and frequencies, as well as with spike triplets.

Submicron Ambipolar Nanocrystalline Silicon Thin-Film Transistors and Inverters

Nanocrystalline silicon (nc-Si) thin-film transistors (TFTs) fabricated at a maximum processing temperature of 250 °C operate with high field-effect mobility compared with amorphous-silicon TFTs. By reducing the oxygen content in the channel layer, ambipolar behavior can be obtained. Two levels of electron-beam lithography are employed to fabricate nc-Si TFTs with nanoscale dimensions that operate without significant short-channel effects for gate lengths down to 200 nm. The TFTs have current-voltage (I- V) characteristics with on-off ratio >; 105 at ±1 V drain voltage and low threshold voltage shift. Simulation Program with Integrated Circuit Emphasis (SPICE) software is used to model the TFTs, and it is validated by performing the fit to devices of different dimensions. An inverter constituent of nc-Si TFTs offers high voltage gain (10-12) and frequency response better than 2 MHz. The crowbar current associated with the inverter can be minimized by using an optimized geometry ratio based on the leakage currents of the TFTs. An amplifier circuit is also demonstrated, offering an ac gain in the frequency range of 100 Hz-10 kHz. SPICE simulations of the inverter and amplifier show close agreement with measured data. The fabricated devices are well suited for use in high-density architectures.

SPICE simulation of nanoscale non-crystalline silicon TFTs in spiking neuron circuits

Electrical characteristics of nano-crystalline silicon (nc-Si) thin-film transistors (TFTs) are fit using SPICE device models. The corresponding device model geometry is then extrapolated down to submicron dimensions using electrical data measured on a-Si:H transistors as justification. The nanoscale devices are then used to simulate a spiking neuron circuit. The frequency of output voltage pulses in the circuit is a function of the input current. Various loads are added to the output to represent driving of many synapses. Frequency versus current curves from the circuit simulations are compared to biological models. The similarities demonstrate the feasibility of using low-temperature, large-area semiconductor materials such as nc-Si in nanoscale devices to implement neuromorphic electronic designs.

Hydrogenated amorphous silicon nanowire transistors with Schottky barrier source/drain junctions

Hydrogenated amorphous silicon nanowire field-effect transistors (a-Si:H NWFETs) with Schottky source/drain junctions have been fabricated with a simple process involving maximum temperatures of 250 °C. Electrical characteristics of devices with various numbers of wires and different linewidths are analyzed. The NWFETs with small effective channel width demonstrate improved subthreshold slope and field-effect mobility as compared to wider devices. Additionally, the on-current scales linearly with effective channel width. Possible explanations for these effects are discussed, and applications of a-Si:H NWFETs are presented.

Spike timing-dependent synaptic plasticity using memristors and nano-crystalline silicon TFT memories

Neural circuits based on ambipolar nano-crystalline silicon TFTs and memristive synapses are investigated via SPICE simulations. The drive transistor for the memristive devices is an ambipolar TFT with memory that could be physically implemented using a metal nanoparticle layer within the gate dielectric. It is shown that using such a device adds spike-timing dependence to changes in the synaptic weight. In experiments with action potential pairs, the synaptic weight modification is similar to biological data. Further, asymmetric temporal integration of the weight change is demonstrated using pre-post-pre and post-pre-post spike triplets. Finally, the dependence of weight changes on frequency is presented. This is followed by a discussion of applications and issues which require further analysis.

Ambipolar nano-crystalline-silicon TFTs with submicron dimensions and reduced threshold voltage shift

Hydrogenated nano-crystalline-silicon (nc-Si) thin-film transistors (TFTs) are primary candidates for use in neuromorphic circuits and systems [1]. Such devices can be fabricated at low temperatures and over large areas, allowing cheap processing and three-dimensional integration with CMOS structures. The major drawbacks of nc-Si TFTs include low carrier mobility, threshold voltage (VT) shift under bias stress and lack of p-channel operation due to unintentional n-type doping by oxygen impurity present in the nc-Si layer [2]. We have fabricated nc-Si TFTs that minimize all the above drawbacks, and are thus well suited for use in neuromorphic applications.

Submicron ambipolar nanocrystalline-silicon TFTs with high-K gate dielectrics

Thin-film transistors (TFTs) that use hydrogenated nanocrystalline-silicon (nc-Si) as the channel material are favorable for use in neuromorphic circuits [1] as well as in flat-panel displays. Nanocrystalline-Si can be deposited over large areas at low temperatures, thus enabling three-dimensional integration with CMOS structures. Recently, nc-Si TFTs that exhibit high channel mobility and provide stable operation under voltage bias stress have been fabricated [2]. In this work, the subthreshold swing, electron and hole threshold voltages (VT), and field-effect mobilities are considerably improved by using high-K dielectrics instead of SiO2. These gains will translate to circuits with lower operating voltages at the same performance.

Logic Gates and Ring Oscillators Based on Ambipolar Nanocrystalline-Silicon TFTs

Nanocrystalline silicon (nc-Si) thin film transistors (TFTs) are well suited for circuit applications that require moderate device performance and low-temperature CMOS-compatible processing below 250°C. Basic logic gate circuits fabricated using ambipolar nc-Si TFTs alone are presented and shown to operate with correct outputs at frequencies of up to 100 kHz. Ring oscillators consisting of nc-Si TFT-based inverters are also shown to operate at above 20 kHz with a supply voltage of 5 V, corresponding to a propagation delay of 5 V for several hours.