Temporal-encoded 6T-RRAM with Bidirectional Control for Future Neuromorphic Systems

Kang Jun Bai1, Hao Jiang2, Zhuwei Qin2, Clare Thiem1
1Air Force Research Laboratory, 2San Francisco State University


Neuromorphic systems are of highly importance in artificial intelligence (AI) to eschew challenges inherent to deep learning acceleration in conventional systems. Throughout the development history of neuromorphic systems, analog in-memory computing with emerging memory technologies, such as resistive random-access memory (RRAM), offer advantages by executing operations in situ, exactly where the data are located, providing significant improvement in data throughput and energy efficiency. Herein, we introduce a temporal-encoded memristive crossbar that offers a faster yet more efficient in-memory operations to accelerate deep learning algorithms. Specifically, the memristive crossbar is built based upon the hafnium-oxide RRAM in a 6-transistor-1-RRAM (6T1R) structure, controlled by a series of temporal-encoded voltage pulses. With a bidirectional switching mechanism, the 6T1R cell can be programmed to support multi-bit operation and represent either positive or negative weight values per single cell. More importantly, the 6T1R cell can be set directly to the high-impedance state, either representing zero or quarantining faulty RRAM from the network regardless of its resistance state. Proof-of-concept simulations are conducted using a custom 65 nm CMOS/RRAM technology node, yielding an average classification accuracy of 94%, 95%, and 99%, respectively, when tested with the banknote authentication, balance scale, and iris plant dataset. Better yet, the parallel inter-tile communications in analog achieves up to 1.6 trillion-operations per second.