Analog-to-Stochastic Converter Using Magnetic Tunnel Junction Devices for Vision Chips

TL;DR

The paper tackles the challenge of converting analog signals from vision sensors directly into stochastic bit streams without bulky ADCs and digital-to-stochastic interfaces. It proposes a magnetic tunnel junction (MTJ)–based analog-to-stochastic converter that uses the MTJ's probabilistic switching to map analog input to stochastic probability, achieving a linear relationship by tuning the write-time and bias parameters. The authors analyze MTJ variability and present a calibration-based compensation strategy (adjusting the write attempt time $t'$ and bias $V_{bias}'$) to preserve linearity, validating the approach with transistor-level NS-SPICE simulations and a simple stochastic vision-chip model in MATLAB. The results suggest substantial potential for area- and power-efficient stochastic vision chips in cognitive systems, with demonstrated robustness to soft errors and a clear path to practical calibration and integration.

Abstract

This paper introduces an analog-to-stochastic converter using a magnetic tunnel junction (MTJ) device for vision chips based on stochastic computation. Stochastic computation has been recently exploited for area-efficient hardware implementation, such as low-density parity-check (LDPC) decoders and image processors. However, power-and-area hungry two-step (analog-to-digital and digital-to-stochastic) converters are required for the analog to stochastic signal conversion. To realize a one-step conversion, an MTJ device is used as it inherently exhibits a probabilistic switching behavior between two resistance states. Exploiting the device-based probabilistic behavior, analog signals can be directly and area-efficiently converted to stochastic signals to mitigate the signal-conversion overhead. The analog-to-stochastic signal conversion is theoretically described and the conversion characteristic is evaluated using device and circuit parameters. In addition, the resistance variability of the MTJ device is considered in order to compensate the variability effect on the signal conversion. Based on the theoretical analysis, the analog-to-stochastic converter is designed in 90nm CMOS and 100nm MTJ technologies and is verified using a SPICE simulator (NS-SPICE) that handles both transistors and MTJ devices.

Figures (19)

• Figure 1: Example of a vision chip used in a cognitive system.
• Figure 2: Spin-transfer torque (STT) MTJ device: (a) stacked on a CMOS layer and (b) its R-I characteristic.
• Figure 3: Probabilistic behavior of MTJ device: (a) write current ($I_w$), (b) switching probability ($p_w$), and (c) switching behavior. The MTJ state is probabilistically changed to another state.
• Figure 4: Multiplier based on stochastic computation designed using a 2-input AND gate.
• Figure 5: Analog-to-stochastic converter: (a) conventional and (b) proposed. The proposed converter directly converts analog to stochastic signals to mitigate the signal-conversion overhead of the conventional system.