Beijing -- Intel Corp. is preparing to sample a 128-Mbit phase-change memory that will roll into volume production in the second half using 90-nm technology.
The device, code-named Alverstone, is Intel's first phase-change memory and is being billed as a NOR flash replacement. Intel is the second-largest NOR flash vendor, behind Spansion Inc.
Intel unveiled the part last week at its spring developer forum here. Chief technology officer Justin Rattner said the part has six times the write performance of today's NOR flash and is much more "robust," lasting for at least 1 million write cycles.
The nonvolatile memory technology is based on the electrically induced phase change of chalcogenide materials, which have been difficult to make reliably in volumes. Phase-change materials have both crystalline and noncrystalline states that can represent "0" or "1," and it's possible to toggle between them by applying a small reset current.
Phase-change memory--also known as PCM or PRAM--is vying with a handful of other next-generation alternatives, such as magnetic RAM, to replace DRAM and flash once they are tripped up by scaling limits. Many technologists expect flash to reach its limit at 45 or 35 nm.
But uncertainties still dog options like MRAM. It doesn't scale well in terms of density, with current chips hovering around 16 Mbits, said Celeste Crystal, a semiconductor analyst at IDC. "The 128-Mbit PCM technology [Intel] announcement is significant in that it shows that PCM will soon be viable in a density that fits today's application requirements. I have also heard that it is backward compatible to NOR sockets. The main thing to keep an eye on is if Intel is able to execute to plan," she said.
A typical handset has 256 Mbits of discrete NOR flash, or about 128 Mbits in a cost-sensitive phone. Rattner said Intel has set "modest" goals by targeting the 128-Mbit part as only a NOR replacement. But he added, "It does demonstrate the performance potential of phase change."
Flash still has legs
It's too early to tell which memory will succeed flash, said Jim Handy, princi- pal flash analyst at Objective Analysis: "Although all these technologies--PRAM, MRAM, FRAM [ferroelectric RAM]--have been touted as the replacement for flash once it reaches its scaling limit, that limit keeps being pushed out."
Handy believes that until flash reaches its limit, these technologies will be in catch-up mode.
Rattner said Intel will not only use this first-generation part as a product, but also as a way to fine-tune the phase-change memory mass-production process. "If it can be manufactured in high volume and at low cost, it will lead to a lot of rethinking of the memory hierarchy," he said. "It is fundamentally cheaper than DRAM, and if it performs well as a fast read-write memory with nonvolatility, then it makes a pretty compelling DRAM replacement."
Last summer, Intel joined forces with STMicroelectronics NV to develop phase-change memory. The companies jointly presented research at the VLSI Technology Symposium. Although Intel is moving ahead with mass production, ST doesn't expect to do so until it ramps up the 45-nm process node. ST currently has a 128-Mbit large-area demonstrator imple- mented on 90-nm process technology.