Work

Integrated ZnO-Based Synaptic Devices to Implement Multiple-Layer Neural Networks on CMOS Circuit Platform for High-Performance Synapse Applications

Year: 2026

Type: article

Abstract: ZnO-based synaptic devices employing large-bandgap materials─HfO 2 /ZnO and ZrO 2 /ZnO─have been proposed for high-performance synapse applications. In this study, bilayer HfO 2 /ZnO and ZrO 2 /ZnO synaptic devices were fabricated on an 8-in. complementary metal-oxide semiconductor (CMOS) integrated circuit platform to address the key challenges of synaptic devices, including their variability, reliability, biological synaptic functions, and integration into neuromorphic chips. A 64 × 64 (4096 devices) one-transistor, one-resistor (1T1R, R: synaptic device) array chip with DEMUX and input/output circuits was fabricated, and the synaptic devices within the array were thoroughly analyzed. Leveraging CMOS technology, the array exhibited ±1.4% variability in selection transistors, confirming that the intrinsic characteristics of the synaptic devices could be evaluated without interference from the selection transistors. Moreover, the bilayer synaptic devices, integrated on 8-in. wafers, demonstrated excellent recognition accuracy, supported by the linearity and symmetry of their long-term potentiation/depression characteristics. Notably, a critical feature for practical utilization in 64 × 64 (4096 devices) 1T1R array chips was achieved by modulating the recognition of input patterns. An analog synaptic device neural network achieved an accuracy comparable to that of state-of-the-art digital CMOS systems, with the potential for major improvements in speed–energy efficiency for high-performance spiking neural network and deep neural network applications.

Source: ACS Applied Electronic Materials

DOI: 10.1021/acsaelm.5c02491