In this study, we propose a modified predictive direct torque control (PDTC) application-specific integrated circuit (ASIC), comprising a neural network (NN) proportional integral derivative (PID) controller, speed-sensorless control, fuzzy error controller, and seven-stage hysteresis controller, to alleviate the ripple problem induced by limited vector voltages and slow speed response in conventional direct torque control. Both flux and torque errors pass through the modified discrete multiple vector voltage switch table to obtain the required vector voltages, and the proposed NN PID controller is used to convert the speed error into a torque command. Notably, the motor speed is evaluated from the magnetic flux, which is calculated using two-phase currents and voltages. The speed-sensorless control not only accelerates the feedback control but also rotates more stably. The NN PID controller generates a torque command according to the speed error, which is obtained by subtracting the estimated predictive speed from the actual speed. The advantages of the proposed system are that it reduces the flux and torque ripples and increases the control stability by filtering out the external interferences. The Verilog hardware description language is used to implement the proposed PDTC ASIC system, and a field-programmable gate array development board is used to verify the designed functions.