Homogeneous photoelectric reservoir computing system based on chalcogenide phase change materials

A neuromorphic visual system integrating photoelectronic synapses to perform the in-sensor computing is triggering a revolution thanks to the reduction of latency and energy consumption. Phase change materials based on Ge-Sb-Te ternary alloy have become a strong candidate for neuromorphic computing due to its compatibility with complementary metal oxide semiconductor (CMOS). Hence, a homogeneous photoelectronic reservoir computing (RC) system based on chalcogenide phase change material is proposed in this work. The reservoir and readout layers are realized by the same material, and the sign language recognition is implemented by in-sensor computing and in-memory parallel computing. By doping N into Ge₁Sb₄Te₇ (NGST), the conductance modulation linearity, symmetry and retention of the phase change electrical synapse are improved, making the NGST electrical synapse excellent for readout layer. Meanwhile, the nonlinear optical response characteristics and persistent photoconductivity (PPC) effect of amorphous-NGST (a-NGST) enable the a-NGST photo-synapses to form an ideal photoelectric reservoir. The system's sign language recognition accuracy can reach 99.58 %. With a random noise level of 15 %, the system's sign language recognition accuracy remains above 90 %. This homogeneous design for photoelectric RC system shows excellent process compatibility and high integration. Furthermore, due to the excellent retention characteristics of the NGST synaptic device in the readout layer, the system's sign language recognition accuracy remains 97.60 % after 10⁶s. This work shows that the chalcogenide phase-change materials have great potential in in-sensor computing applications.