- Aug 5, 2002
- Silicon Valley
You can ask the OCZ guys why they can't do it. The contract manufacturer I work with told me 2nd/3rd hand info that they bought whole wafer from Toshiba and do their own testing to save a few bucks, but they don't know what they were doing (how to stress test it) or cut some corners. Went out of business and sold their brand to Toshiba. Every NAND chip has bad block marked during testing and over provisioned for reassign / reallocation. The problem is it cost money to mark the blocks bad too aggressively (in wasted capacity), or not (in warranty claim). One aftermarket brand guy told me that when they evaluate controller vendor, they look at the relationship with NAND manufacturer too, because that's how they know how to screen and work around design weakness (i.e. write in certain sequence or test in a certain way to screen for certain recently found problem).
Originally Posted By: emg
Bad chips can be worked around, though. The first SSD I used many years ago was built from a complete RAM chip wafer, and they wired up all the good chips and ignored the bad ones. It had zero data retention because all the data was lost if the power went out . I believe some modern SSD manufacturers do something similar, by installing more flash than they need and testing to see which blocks are good and which are bad, then only using the good ones. We shipped about 200 enterprise SSDs to customers in the last year and have only had one fail so far (sudden death, so probably a firmware bug). At home, I upgrade the SSD in my laptop to a new one of double the size every couple of years, then use it to replace the SSD in my Windows desktop, and use that one to replace the SSD in my Linux desktop (which has no essential files). Haven't had any problems so far, though this 1TB Kingston is the first non-Intel SSD I've used. Back more on topic, I've never had problems with laptop HDDs other than the Toshiba drive that came in the last laptop, which failed a couple of weeks after the warranty expired. But, even though there was a rapidly-increasing number of bad blocks, I was still able to get all the data off it.