A Multilevel Write-Once-Read-Many-Times Memory Array for Convolutional Operation

Write-once-read-many-times (WORM) memory, as a key nonvolatile memory technology, offers significant advantages for permanent and low-cost storage applications. However, it currently struggles to achieve multilevel memory behavior, which is crucial for future high-density data storage and analog computing applications. In this letter, we report a multilevel resistive WORM memory based on solution-processed polyimide (PI)/carbon nanotube (CNT) composites. The memory leverages a controlled carbonization process to achieve multilevel WORM memory characteristics, resulting in an eight-level distinct permanent conductance state. Moreover, we fabricated a multilevel WORM memory array consisting of 18 WORM devices, which constructed a Laplace convolution kernel for feature extraction operations. Experimental results align closely with simulation results, indicating the analog computing capacity of the WORM memory. The findings underscore the potentials of WORM memory in analog computing technologies and edge intelligence devices.