Log in

View Full Version : hard errors and soft errors in memory


eclectica
2003-12-16, 02:24
Memory errors occur occasionally in computers. Errors are when a zero should be a one, or the opposite. Computers could crash unless they are using error correcting memory chips. A simple type of error correction is parity checking, which will add up the eight ones and zeroes in a byte, seeing whether the result is correctly even or odd. But in the unlikely event of there being two errors, the error won't be caught with parity checking. More advanced is something known as ECC memory, which will detect more than one bit error per byte.

There are two types of errors: hard errors, and soft errors. Hard errors are rare occurences due to a physical defect, and will continue, so that the memory chip must be thrown out. Soft errors are caused by electromagnetic radiation and are not permanent. By various estimates, soft errors occur a few times a year with 128 MB of memory at sea level. At high airplane elevations, they occur 100 times as much. That is because there is less of the atmosphere protecting users from cosmic rays at higher altitudes. The likeliness of failure is also proportional to the amount of memory in a computer. 256 MB of memory makes a computer twice as vulnerable as 128 MB of memory. So figure if you're in an airplane with a new laptop, you have a good chance of getting a soft error and having your computer crash. ECC memory is more expensive and a little slower, but it is a worthy investment for high end computers and servers. Most computers don't have it.

"Moore's Law" is an observation that chip data density doubles approximately every eighteen months. But the problem with smaller chips, is that they are more susceptible to radiation, so Moore's Law is no longer holding true.

STMicroelectronics (http://www.st.com/) has announced a new chip it designed which it boasts is 250 times less likely to be hit by radiation and soft errors. Their design involves changing the physical layout of the chip. It is thicker and bunkered in by a capacitor on top of it. If their design proves successful and reliable, it could be a way to produce more dense chips and for the predictions of "Moore's Law" to continue to be accurate.