A design method and implementation of a programmable cellular neural network for optimization and image processing applications is presented. The time-varying cell gain (TVCNN) “hardware annealing” is also embedded in the network. The test of such a system showed highly efficient in finding globally the optimal solutions for cellular neural networks. The cell gain as an annealing control signal is implemented by using a continuously adjustable MOS amplifier. The adjustable amplifier that used for this function combines an active input and a regulated cascade output. The proposal design method of TVCNN will be implemented by applying both the voltage-mode and current-mode concepts to have an idea of cost and speed in our implementation decision. Experimental simulation shows that the proposed approach is effective for real-time signal and image processing using standard CMOS technology. It offers a high accuracy over an input range. The analytical formulas for determination the values of designable parameters as a transistor sizes and component values are illustrated by an example of optimization task.