Del,
Before yesterday I would have found it difficult to agree with you, but I never totally discounted the possibility, because, as you say, my mind is not the best one in the business.
Your statement because it is so desirable to have is very true, but I think what is even truer is that the huge dynamic range at a single temperature makes it so tempting to find a way to harness it.
My multibit-skepticism was a product of 3 things:
1) Programming resistance HIGHLY sensitive to temperature, as evidenced by the charts. The movement with temperature is huge!
2) The extra complication of variations in temperature from point to point on the chip because of variance in local read/write activity.
3) I had thought that the resistance of a programmed cell would change with temperature, so that you could program a cell to one resistance and then later that resistance would be different if the temperature changed. This is different from #1.
But, then, it occurred to me that #3 might not necessarily be true. With this possibility in mind, the job of rectifying the whole messy affair seemed much simpler and...... POSSIBLE.
Of course, once you conceive of rectifying the writes, which effectively restores the programming curve to its 1-temperature pristine-ness, even a messy drifting of (post-write, or programmed) resistance with temperature seems not so formidable, because it becomes a single separate problem -- in isolation.
4 levels/2 bits affords a LOT of margin for error because of the huge dynamic range, which is about 35X on the 80C mil-spec curve. It allows for 200% gaps between resistance levels:
1.00 3.00 9.00 27.00
8 levels/3 bits allows a 65% gap:
1.00 1.65 2.72 4.49 7.41 12.23 20.18 33.30
16 levels/4 bits allows about 25% (you'll have to trust me).
How many bits/cell they can achieve just depends on the degree of programming precision. I'd be very happy with 2 bits/cell, as it would give OUM a large enough cost advantage to drive rapid market penetration. Multibit truly gives OUM, as Tyler has said, an UNFAIR ADVANTAGE.
With 2-bit/cell OUM, Ovonyx would need only 2 Flash licensees: Intel and AMD/Fujitsu (FASL). Their present shares of the Flash market are 25 and 30 percent respectively. Double their capacity by converting to multibit OUM, and their market coverage could be 110%. People on the Sandisk thread like to think of Flash as disruptive technology -- I guess they don't realize what shaky ground their earthquake is sitting on.
wily |
