Bias is an offset that occurs when a pixel is read from the camera. Unfortunately, bias can vary across the image. Also, if bias is not corrected then flat-field calibration will not work correctly.
A bias frame is essentially a zero-length exposure (or as close as possible to zero length) with the shutter closed. Each pixel will have a slightly different value, but except for a small amount of noise, the value for any one pixel will be consistent from image to image. Since the bias is consistent from image to image it can be subtracted.
The bias frame itself contains a small amount of readout noise. This readout noise is produced inside the electronics that read the pixels. It can be very low in sophisticated cameras, but it is never zero. This noise can be easily suppressed by combining a number of bias frames together.
Ideally, the other types of calibration frames (dark and flat-field) should themselves be bias frame calibrated. MaxIm DL does this automatically when bias frame files are selected.
The bias for a particular camera is generally constant over a substantial period of time. This means that you can take bias frames just once, and use them on all your images for many months to come. Note that some cameras may have a small bias dependency on temperature. Small bias offsets are not important of themselves, but they can degrade the effectiveness of the flat-fielding calibration.
It should be mentioned that bias is also included in dark frames. As such, it actually is possible to perform accurate calibration without using bias frames. That said, you should always use bias frames if either of the following are true:
You are using exposure scaling to match dark frames to light frames
You are using flat-field frames but are not using matching dark frames ("flat-darks")
That said, bias frames are extremely easy to acquire, so there is little reason to skip them. You can quickly acquire 10 or 20 frames to average, so the read noise in the bias frames will not be contributing to your overall noise level.