Adaptive immunomorphic hardware based on organic semiconductors and oxidized MXene heterostructures for feature information recognition

Abstract

Neuromorphic computing with brain-like functions has become one of the important strategies for the von Neumann bottleneck. However, current artificial neuromorphic systems are mainly based on volatile or non-volatile synaptic devices, which limit the flexibility and computational efficiency of neuromorphic computing systems. Here, we report an adaptive immunomorphic hardware based on the heterostructure of MXene-TiO 2 complexes and organic semiconductors. The hardware has photon-triggered synaptic plasticity for accurate recognition and electrically triggered non-volatile retention for effective preservation of weight values. As a result, the retraining time and power consumption of the hardware can be reduced by 95% and 96%, respectively. Moreover, the array system expanded to 5 × 5 can extract special information from complex signals within 0.2 s, enabling feature information recognition. This work, therefore, provides a new strategy for improving the efficiency of artificial neuromorphic computation and has significant application prospects in intelligent sensing systems and edge computing. • Photoelectric transistor with memory window and memory ratio exceeding 60 V and 10 5 • Combination non-volatile and volatile retention for feature information recognition • The power consumption and time were reduced by 96% and 95% in retraining Gao et al. report a photoelectric transistor with the heterostructure of MXene-TiO 2 and P2FDIID. The device can achieve non-volatile and volatile retention behavior under electrical and optical pulse, respectively. As a result, the combined operation of the two retention characteristics can significantly reduce power consumption and time in retraining.