Your general pricing remarks (trends) are obviously valid. Data storage is growing fast for cell phones. But OneNAND and mDoC only provide the cheaper solution when data needs are very big / high end phones (because often more total NAND/SRAM/etc-die area is needed for equal performance). And if current pricing trends do continue long enough, i.e. two years, and if NOR technology would not continue to evolve rapidly then all current NOR will indeed fade away. Still, because of the technical difficulties of scaling NAND at an equal pace as currently all flash manufacturers (NAND and NOR) are changing technologies. E.g. Spansion bet on charge trapping a long while ago, while Samsung is doing something more or less similar now (Numonyx didn't innovate nearly as much). Second example: Both Spansion and Toshiba are implementing SONOS (similar cell sizes for NOR and NAND I/F's); Spansion when entering the 45nm node, and Tosh at least one generation later. In the end (years from now) there might simply be little technology difference between NAND and NOR on memory cell level; the difference will be in the layout and the I/F. If Spansion does execute current faster roadmap after 45nm too then they're in a good position to come out of the transition as the only previous NOR manufacturer still standing and possibly even in good shape. Samsung rather obviously too of course. Not Numonyx though because they're late in shifting memory cell technologies and late with 300mm production.
NOR will not evolve rapidly because the market is not growing for them and poor RoI does not justify enormous investments. The amount of funds that are being poured into NAND is a clear indication that semiconductor manufacturers are targeting NAND market for explosive growth for at least next 5 years! NOR industry is in overcapacity and market is not growing so C/Fs are not there to finance the road map. Spansion does not have the necessary resources to continue on an aggressive transition path!
Almost 100% true. 'Almost' because Spansion's just started sampling 4 bits / cell 65nm flash using 300mm wafers that has a cell size comparable with 2 bits / cell NAND at ~50nm. OneNAND die is significantly bigger than standard NAND because of the additional SRAM buffer and NOR I/F + error correction.
It is not "almost" 100% true, it is 100% true. NOR flash cells have a feature size of 10f^2 and NAND flash cells have a feature size of 4f^2. NOR with 4 b/c will have an effective cell size of 10/4=2.5f^2 whereas NAND with 2 b/c will have an effective cell size of 4/2=2f^2! In any case, you are comparing apples to oranges. You should look at cell size if both are at the same node to get an insight about cost/MB. Since NOR is at least 2 generations behind, even if they had a smaller effective cell size, the cost of memory per MB would be much higher than for NAND, which is at a more advanced node and the bottomline is cost/bit is what matters! If NOR and NAND are at different nodes, you should be looking at the die size for comparable capacities. Of course, one must consider other issues also, but I will not go into the details. NOR is lot more expensive compared to NAND and will remain so for the foreseeable future. To make long story short- NAND can do what NOR has been doing for years at a much lower price but NOR cannot do what NAND can do as it was not designed for data storage!
Essentially true. Some minor details though: Tosh delayed sampling of 3 bits/cell NAND from end last year to somewhere during H1 '08. 3 bits/cell performance isn't good for code, so it will for a good amount of time only compete with NAND where costs are more important than performance. Die size improvement (I believe I once read 20% reduction) is good but not dramatic because of added die area to compensate for worse reliability (EC).
You are right, Toshiba did delay 3 b/c but it is not clear whether it was a planned or an unplanned delay. These bits may not be good for code execution but majority of the growth for NAND is coming from data storage applications and not from code execution opportunities. Bit requirements for code execution is a relatively small market and will remain small when compared to data storage market for obvious reasons.
Spansion has been in production with 65nm ORNAND (2b/cell) since September last year. 1Gb announcement is just for Quadbit (initial sampling density for 65nm Quadbit). Spansion's mainstream flash 2b/cell Mirrorbit NOR hasn't been sampling yet though fafaik. Spansion is moving to 45nm end this year.
Yes of course NAND will scale, but probably not as fast as currently because process technology changes are imo likely to slow it down. Noone will want to move to 3D transistors unless absolutely necessary. That's why they've been delayed time after time so far.
With Mirrorbit (NOR) / ORNAND (Spansion's NAND) / Quadbit Spansion wins out nicely over Numonyx. With products like OneNAND and Flex OneNAND Samsung wins business from Numonyx too. Samsung 'only' wins high end cell phone NOR share from Spansion; I'd be rather curious to what extent this is the case btw.
First of all, we should not even compare NOR and NAND. They are two different types of NVMs and more importantly, NOR is not designed for data storage. By the same token, NAND has not been designed for code execution but it so happened that NAND is getting better with R/W speeds and some smart people figured out how to get NAND to do what NOR has been doing for years, with some SRAM and a controller, at much lower price points! The reverse of this, NOR mimicking NAND, is not going to happen if you understand how NOR works. So even if the pace of scaling slows down for NAND, NOR will not be competitive with NAND for the next 5 years at minimum, if at all!
Spansion too has highly depreciated fabs.
Spansion has no chance to compete against NAND flat out, but does have a good chance with NOR+NAND MCP's especially where data needs are not very big.
Spansion may have depreciated fabs but unfortunately, they do not have the diversification that others have! Nor do they have sufficient resources to fight the price war that is looming in the NOR industry. Spansion has no chance in NOR+NAND MCPs either beyond short-term. Where is the synergy? Very little NOR is needed but a lot of NAND is needed in a MCP solution, and their NAND cost will never come close to tier-1 NAND players' cost per MB over the next 5 years at minimum! In any case, they also need a controller for the NAND portion in their MCPs.
Your price remark is true only in very high end. It can become true in lower segments too
Spansion is indeed too late to the NAND party, but not the NOR+NAND MCP party.
Spansion does indeed need to worry about survival (see share price) and will indeed not be profitable in 2008, though they could reach break even operating profitability that year. Also they will continue to burn cash at moderate levels because of SP1. 2009 is less clear but the trend is ok currently.
Partly in line with you I think the NOR market will remain essentially flat for some time and shrink over long term. Numonyx does not have a single fundamental chance to compete (or it must be highly successful phase change; I have to see that to believe it). Competition will indeed be fierce but Numonyx can't lower prices the same way Intel did. Essentially Samsung and Spansion will determine NOR pricing. In general making money in NOR will only get harder, very true.
We agree on outlook for NOR but the market is disagreeing with your assessment of Numonyx. The deal is not done yet so it is a bit premature, but the market is currently valuing Numonyx much higher than it is valuing Spansion. In any case, we are digressing and the bottomline is NOR market will be stagnant or shrink so if some is looking for growth, they are in the wrong place ;-) |
