ICM-42688-P raw data drift

Hello, everyone! I've been programming with an ICM-42688-P for its data of acceleration and angular velocity, and now something wrong happens. The output of gyroscope vibrated between -255 to +255 (scale of 2000, 8000 HZ sampling) while it stayed stationary. Sometimes the output data comes back to near 0, but it will jump to ±255 in just a few epochs. Meanwhile, the 3-axis gravity output of accelerometer in rest condition reached 9.87 m/s2, appearing a giant constant zero bias.
How can I deal with these problems? Hope for help!