Unlike flash memory, the BIOS settings non-volatile memory (which has, at least historically, often been implemented using complementary metal-oxide-semiconductor technology, hence the "CMOS memory" nickname) gets cleared if power is removed from it. A small battery connected to the motherboard provides power when external power is removed from the motherboard. That battery also drives the real-time clock when other power sources are unavailable. Like all batteries, eventually it gets depleted and needs to be replaced; this can be expected to happen after 5-10 years, depending on system usage. "CMOS checksum error" (and similar messages) or the system clock being reset are classic signs of the battery having been depleted.
The BIOS settings contain several system-critical values. When the memory is no longer maintained because the battery has run out, everything reverts to a pattern that is determined by the physical and electrical design of the memory, as opposed to a pattern matching the settings entered by the user. A mechanism needs to be in place to detect and handle this situation.
One could imagine the settings memory being scanned and some set of sanity checks being performed on the data, which would probably catch the above in the vast majority of cases. But it might not catch all of them, and a checksum provides additional benefits. Primarily, it allows catching any inadvertant corruption of the data. Remember that in the original IBM PC XT/AT design, any process could write to any memory location, including issuing write instructions to the BIOS settings memory, so an errant piece of software could (inadvertantly or maliciously) corrupt these settings. However small, there is also the possibility that a legitimate rewrite of system settings could be interrupted, for example due to a power failure.
The exact mechanics of how to compute the checksum, where to store the checksum, what exactly is covered by the checksum, and so on, are implementation-specific for each particular BIOS and quite possibly BIOS version. The important part is that it needs to detect the case where the memory has been reset. To cover for different memory chip designs, this means it needs to be more sophisticated than a simple sum (modulo 2^n) of all bytes, but at least originally, IBM's ROM BIOS was 8 KiB (which needed to do a lot more than this) so it can't have been a particularly complex algorithm either.
Since the data protected by the checksum is partly system critical, simply ignoring the error (which really I wouldn't call an error; it's more of a warning) would be inadvisable. Instead, you should enter the configuration utility ("BIOS setup") and perform whatever ritual dance is necessary to perform an autodetection or reset to factory configuration, then adjust whatever settings need to be adjusted from the default values for your particular system. Some systems which knew that the data was non-critical might have ignored a checksum mismatch and carried on, but in the IBM PC/AT onwards, doing so would mean you were at risk of using incorrect system configuration data (including hard drive geometry; HDD geometry autodetection came much later, and I recall using a 386-based system which required manual HDD geometry configuration through the BIOS setup).
