Novel a-IGZO Anti-Ferroelectric FET LIF Neuron with Co-Integrated Ferroelectric FET Synapse for Spiking Neural Networks
Chen Sun, Xiaolin Wang, Haiwen Xu, Jishen Zhang, Zijie Zheng, Qiwen Kong, Yuye Kang, Kaizhen Han, Leming Jiao, Zuopu Zhou, Yue Chen, Dong Zhang, Gan Liu, Long Liu, Xiao Gong
Abstract
For the first time, a novel amorphous-Indium-Gallium-Zinc-Oxide (a-IGZO) anti-ferroelectric field-effect transistor (AFeFET)-based leaky integrate-and-fire (LIF) neuron is experimentally demonstrated, emulating both excitatory and inhibitory input connections with capacitor-free neuron design. By co-integrating a-IGZO ferroelectric field-effect transistors (FeFETs) as synapses, spiking neural networks (SNNs) with high biomimetic and low hardware costs could be implemented. The highlights of this work include: (1) high-performance AFeFET with channel length $(L_{CH})$ down to 50 nm and endurance of more than $10^{9}$ cycles is realized; (2) the inherent volatile feature of AFE HfZrO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> (HZO) and ferroelectric dynamic switching offer the flexibility to leverage the leaky and accumulation effects by adjusting the base voltage $(V_{B})$ of input pulses; (3) a-IGZO AFeFET neuron and non-volatile FeFET synapse with the same metal-ferroelectric-metal-insulator-semiconductor (MFMIS) structure and optimized memory window (MW) are successfully integrated; (4) using the experimentally calibrated neuron and synapse models, an unsupervised SNN employing the spike-timing-dependent plasticity (STDP) method is simulated, achieving 91.4% accuracy in digit recognition.