MADISON, Wis. — Globalfoundries this week rolled out a 55-nm semiconductor manufacturing platform, specifically designed to meet the stringent needs of the automotive industry.
The foundry's new automotive platform supports the implementation of non-volatile memory (NVM) in MCUs and SoC designs. Globalfoundries is also making available, on the same 55-nm platform, SST's embedded flash technology -- based on the SST-Microchip SuperFlash split-gate design. This opens a licensing door to second- or third-tier automotive chip suppliers with no flash technology of their own.
Cerntainly, there are other foundries also offering SST technology on their automotive qualified platforms. But Globalfoundries believes it's the first to make it available on 55-nm.
The move toward licensable NVM technologies reflects "the changing automotive foundry landscape," Paul Colestock, director of segment marketing at Globalfoundries, told EE Times.
Why proprietary NVM?
Looking back, many leading automotive chip suppliers resisted going fab lite, let alone going fabless, long after other parts of the same semiconductor manufacturers embraced the fabless model (for example, in making leading-edge apps processors for the mobile market), Colestock observed.
But since the 90-nm process hit the automotive chip market (around 2010), tradition started to crumble: integrated device manufacturers (IDMs) began working with foundries. Colestock calls this "phase one."
During that period, IDMs transferred to foundries core process technology that had been developed in-house. But IDMs continued to work on proprietary embedded memory technologies, and kept them close to the vest.
Why proprietary NVM? Embedded MCUs used in automotive, embedded NVM -- which needs to hold data and code -- must be highly reliable. This is so critical that many automotive chip vendors believed that they had to differentiate their embedded MCUs from everyone else's.
In "phase-two" era, where the industry is in today, Colestock explained that IDMs no longer develop their process technology themselves. They have yielded to foundries. Moreover, IDMs today are more willing to work with their contracted foundries on their proprietary flash technologies. That model allows a foundry to fabricate a certain proprietary embedded flash technology, but only for their IDM customer, not for anyone else.
Colestock sees "phase three" as the destination.
Although still several years away, phase three will happen when foundries offer both the process technology and "industry-standard" embedded flash memory technology to automotive chip vendors, Colestock noted. "Certainly we are not there yet, and we may not get there any time soon," he added.
The industry standard NVM for automotive
Indeed, it's hard to imagine automotive chip companies rallying around "industry-standard" embedded flash memory technology, at a time when practically every leading automotive chip company is armed with its own embedded flash.
As Jim Handy, director of Objective Analysis, told EE Times, "IDMs use a variety of technologies, from the SST-Microchip SuperFlash split-gate design, which is very popular, to Freescale's proprietary silicon nanocrystal flash which is only used in a few Freescale MCUs. One or two IDMs use MirrorBit technology from Spansion."
Alan Niebel, analyst at Web-Feet Research, added to the list a whole lot more. Other embedded NVM technologies he included are:
MONOS embedded flash pioneered by Renesas; SONOS embedded flash used by Cypress; emFeRAM by Fujitsu; Fujitsu's emFlash (Floating Gate) sold to Spansion; emMRAM (qualified for automotive) produced by Everspin; CBRAM (Conducting Bridge RAM) developed by Adesto and offered through Altis, with Panasonic producing their own emCBRAM (although CBRAM may not be qualified for automotive yet); emPCM developed by STMicroelectronics but is also not yet qualified for automotive; Samsung, Toshiba, Atmel, ST, and many others offering emFlash (Floating Gate).
Important in the automotive context is that car OEMs are looking for embedded memory that must be integrated into MCU-MPU. In other words, the memory cells are part of the silicon of another component (MCU/SoC), and aren't packaged as a standalone component.
Next page: 28nm, the battleground