“…Neuromorphic computing, a disruptive computation technology, aims to address these bottlenecks by mimicking parallel information processing and storage in a human brain. , Two solid nonvolatile memory (NVM) architectures, two-terminal memristors − and three-terminal memtransistors, , have been proposed for implementing neuromorphic computing. As the terminals for reading and writing are decoupled, basic memory operations in the memtransistors become nondestructive, which can considerably simplify addressability in crossbar arrays and efficiently enable tunable learning rules and biorealistic functions, e.g., heterosynaptic plasticity, continuous learning, neuromodulation, and multitemporal plasticity. , The most common memtransistor is based on the floating gate (FG) type structure, where charges are stored in the floating gate to modify channel conductivity. − However, gate dielectrics serving as a Fowler–Nordheim (FN) tunnel barrier in the FG structure for hot-electron injection lead to high power consumption, slow programming speed, and poor device endurance. , In contrast, a direct charge-trapping memtransistor (DCTM) is built based on a simple metal–insulator–semiconductor field-effect transistor architecture with a tailor-made band alignment, which demonstrates a promising potential for next-generation memories. − …”