Uncorrectable bit error rate is a term commonly used in memory devices to express the probability of not being able to return data.For SSDs, JEDEC defines it as UBER = (number of data error)/(number of bits read).
In most cases this is just the inverse of the NRRE. I’m not fond of the term UBER, as it implies the errors are associated with reading the individual bits. However, there are many other contributors, such as erase failures, addressing failures, etc. I prefer NRRE.
One caution with the JEDEC definition – they define data errors to include both unrecoverable sector errors (NRRE) and miscorrected data. I have two major concerns with this definitions.
- By definition, the device can’t know it has returned miscorrected data. If it could detect it, it would be an NRRE.
- We shouldn’t combine the miscorrections and NRRE in the same specification. Most RAID controllers can handle NRRE, but few can handle miscorrections. These usually end up as silent data corruptions. Thus they require a separate, and much tighter specification than NRRE.