Biologically Realistic Dynamics for Nonlinear Classification in CMOS+X Neurons

Abstract

Spiking neural networks encode information in spike timing and offer a pathway toward energy efficient artificial intelligence. However, a key challenge in spiking neural networks is realizing nonlinear and expressive computation in compact, energy-efficient hardware without relying on additional circuit complexity. In this work, we examine nonlinear computation in a CMOS+X spiking neuron implemented with a magnetic tunnel junction connected in series with an NMOS transistor. Circuit simulations of a multilayer network solving the XOR classification problem show that three intrinsic neuronal properties enable nonlinear behavior: threshold activation, response latency, and absolute refraction. Threshold activation determines which neurons participate in computation, response latency shifts spike timing, and absolute refraction suppresses subsequent spikes. These results show that magnetization dynamics of MTJ devices can support nonlinear computation in compact neuromorphic hardware.

Figures (4)

• Figure 1: Electrical schematic illustrating two NMOS+MTJ neurons connected by a synaptic weight. Each neuron consists of a magnetic tunnel junction (MTJ), represented by the blue and orange layers, connected in series with an NMOS transistor. The synaptic connection is implemented with a voltage amplifier whose gain represents the synaptic weight $w_{kj}$ that controls how activity from one neuron influences the next neuron. $V_\textrm{in}$ is the input voltage applied to the first neuron, $V_\textrm{out}$ is the output voltage of the second neuron, and $V_\textrm{DD}$ is the supply voltage.
• Figure 2: Architecture of the NMOS+MTJ neural network used for XOR classification. Encoding neurons $A$ and $B$ drive input-layer neurons $i_1$ and $i_2$, which connect to output neuron $o_1$ through synaptic weights. A bias neuron provides a constant input to the network. The XOR truth table showing the bias input and target output.
• Figure 3: Simulation results of rows 1 and 3 of the XOR truth table. (a,e) Input currents supplied to neurons $i_1$ and $i_2$. (b, f) Voltage spikes of neurons $i_1$ and $i_2$. (c, g) Current supplied to output neuron $o_1$. (d, h) Voltage spike of output neuron $o_1$.
• Figure 4: Simulation result of row 2 of the XOR truth table. (a) Input currents supplied to neurons $i_1$ and $i_2$. (b) Voltage spikes of neurons $i_1$ and $i_2$. (c) Current supplied to output neuron $o_1$. (d) Voltage spike of output neuron $o_1$.