The storage density and capacity per die in flash memory keep increasing in the past decades. More flash memory dies are assembled in a storage device and system, which also leads to higher power consumption and more heat generation. Thermal throttling is critical to protect the device from data loss and damage due to overheating. It is triggered after the device temperature reaches a throttling point. During the throttling, the device performance, such as data rate or Input/Output Operations Per Second (IOPS), will be lowered to reduce the rate of heat generation. The restriction on the IOPS depends on how the thermal throttling algorithm is designed and implemented. In this paper, a model predictive control (MPC) based thermal throttling method is demonstrated. The simulation results show that this method can achieve efficient thermal throttling and minimize the performance variation, compared to a traditional PID or bang-bang controller.