Dual integration of Approximate Random Weight Generator (ARWG) and Computation-in-Memory (CiM) is proposed for event-based Reservoir Compu-ting (RC) and Spiking Neural Network (SNN). With evaluation of randomness by Hamming Distance and Hamming Weight, this paper reveals that randomness required by random weights of RC and SNN is much lower than that by physically unclonable functions (PUFs) and random number generators (RNGs). Thus, without generating random numbers externally, proposed ARWG generates approximate random weights of RC and SNN that achieve high recognition accuracy and performs multiply-and-accumulate (MAC) operation. As a result, proposed dual integration facilitates event-based RC and SNN training and saves the circuit area to generate random weights.