Introduction to ECC

What is ECC?

ECC is an acronym for Error Checking and Correcting. ECC is used in several areas of computer operations, but the focus of this paper is on ECC in main memory.
ECC is similar to parity. However, where parity can only detect errors, ECC can actually correct most errors. By correcting memory errors, mission critical computer operations can continue.

What kinds of errors occur in RAM?

In order of likelihood, the most common memory errors are: Single-Bit, Multi-Bit, Column, and Row. Single-bit errors are the most common and are characterized by a single bit of data being incorrect when reading a complete byte or word. A multi-bit error is the result of more than one bit being erroneous within the same byte or word. A single column or row error would appear as single-bit errors in multiple words.

How are these error corrected?

ECC memory uses extra bits to store an encrypted code with the data. When the data is written to memory, the ECC code is simultaneously stored. Upon being read back, the stored ECC code is compared to the ECC code generated when the data was read. If the codes don't match, they are decrypted to determine which bit in the data is incorrect. The erroneous bit is "flipped" and the memory controller releases the corrected data.

Errors are corrected "on-the-fly," and corrected data is rarely placed back in memory. If the same corrupt data is read again, the correction process is repeated. Replacing the data in memory would require processing overhead that could accumulate and significantly diminish system performance. If the error occurred because of random events and isn't a defect in the memory, the memory address will be cleaned of the error when the data is overwritten with other data.

How many extra bits are required for ECC?

By encrypting the ECC code and correcting only single bit errors, very few additional bits are required. Unlike parity, the number of ECC bits doesn't increase at the same rate as the bits per word or data bandwidth. As word size doubles, parity bits double, but ECC bits increase by one. So, if a system uses an 8 bit word, it would need 1 bit for parity checking, but 5 bits for ECC. However, a 32 bit word needs 4 bits for parity or 7 bits for ECC. Increase the bandwidth to 64 bits and 8 bits are required for both parity or ECC. Below is a chart comparing the bits required for different data bandwidths for ECC and parity:

So what?

As you can see, at the 64 bit level, it is equally efficient to use ECC or parity. This is how manufacturers use 36 bit memory modules in groups of 2 or more to create an ECC environment. Compaq's Tri-Flex memory bus is such a system. The Tri-Flex bus uses 4 SIMM groups, for a 128 bit bandwidth. Since the 4 modules have 16 spare bits between them, the 9 bits needed for ECC are easily satisfied. Other systems may use special ECC memory modules, but the cost benefit and availability of industry standard 36 bit SIMMs appeals to designers and end-users.

How is ECC used in the real world?

Systems that use ECC, may use it differently. Usually, when data requires correction, the operating system logs the error and reports the error to the system administrator. Multiple errors may be reported for the same memory location, if the data is read more than once without being replaced by different data. If the same memory location is corrected after a system powerdown, a defect is most likely present in the memory and should be replaced.