Elpida (JP:6665) just announced that their 30nm DRAM process is ramping up, and the release has a bunch of goodies buried in it. 30nm production has been in process for a while now, but the details have not been widely dispersed.
The big news was that the 30nm DRAM ramp at Elpida has begun. That isn’t astounding news, the process announcement last September, shipping samples in December, and volume wafers starting out in January were all much more important. Today, Elpida is basically saying, “The wafers we put in in January are done baking, and all looks well.”
More interesting is that 30nm production will be going full bore in its Hiroshima plant by May, and at Rexchip in Taiwan in Q3. 20% of Hiroshima’s capacity will move to 30nm in Q3, rising to 30% in Q4. Rexchip will ramp even harder with 50% of the fab at 30nm in Q3 and almost 100% by the end of Q4. DRAM is a commodity business, and size = cost. If you can shrink, you can sell for less than the other guy. This isn’t new, and this release says that it hasn’t changed much recently either.
The first parts are going to be a fairly standard 2Gb DDR3, followed by a 4Gb mobile DDR2 (LP-DDR2?) chip, and then a standard 4Gb DDR3 chip. Power use is said to be down 20% over the older 40nm process. Once these parts start shipping in volume, the knee of the DDR3 price curve will drop, and you will see the standard memory size for OEM boxes double. On top of that, since DDR3 speeds are not rising much, the die shrink is seen by users as power savings, not higher clocks. You have to love Moore’s law.
In the end, it looks like Elpida is making progress, the next gen DRAM process is upon us, and things should flip over to a higher density soon. They dodged a bullet in last month’s earthquake, and are putting out new chips on schedule.S|A
Latest posts by Charlie Demerjian (see all)
- Analysis: Is Intel’s Broadwell worth making at this point? - Jul 22, 2014
- Intel dynamically scales core counts for Oracle - Jul 18, 2014
- Microsoft decided to extort Windows 7 users too - Jul 14, 2014
- Intel castrates Broadwell gutting performance - Jul 11, 2014
- Intel delays Broadwell again but technically is on time - Jul 9, 2014