Micro makers weigh up 90nm options
STMicroelectronics has said it will start making ARM Cortex M3 microcontrollers with flash memory based on a 90nm process next year. But NXP Semiconductors, which is looking at moving to a low-power 110nm process is less certain about the value of a smaller process right now for microcontrollers.
ST claimed a limited number of customers would receive samples of the 90nm-based microcontrollers later this year, going into full production next year. Renesas Technology launched microcontrollers with several megabytes of on-chip flash using its own 90nm process a year ago, but wrapped around the company’s own SH and, more recently, the RX processor cores. ST claimed the shift to 90nm would allow for larger, but unspecified, memory sizes and faster clock speeds.
Tsutomu Miki, CEO of Renesas in Europe, said the company aims to sell 90nm-based devices to automotive customers, potentially even for more temperature-sensitive engine-management controllers. He said the use of the smaller process was key to the introduction of the recently launched RX core. “If we introduced it at the 150nm or 130nm level, it probably would have not been so competitive.”
Geoff Lees, general manager of NXP’s microcontroller division, said devices with larger memory sizes can benefit from a move to 90nm. “The sweet spot for 90nm is 1-4MB of flash and more than 256KB of RAM,” he said, but added that the economics of 90nm manufacture do not stack up for many 32bit microcontroller parts.
“We don’t see a way to get tiny die with a 90nm process. The best shrink is on the ROMs and large flash shrinks well but medium-sized flash doesn’t because of all the support circuitry you need. And elements such as the power management unit don’t shrink either,” said Lees.
Another issue is with 300mm manufacturing, which all 90nm processes today use, which favours the production of high volumes of a small number of variants. “As an MCU vendor, you are not dealing with a monoculture. A lot of the business is through distribution [with lower volumes] and we maintain a large number of derivatives,” said Lees.
At the moment, NXP is making many of its M3-based microcontrollers on a dedicated 140nm process operated by the 200mm fab it owns jointly with TSMC. Lees said a possible option is to move to a 110nm process at the same fab, potentially using a similar ultralow-power (ULL) process similar to the one introduced by the Taiwanese foundry for its 180nm fabs. NXP uses the 180nm ULL for its recently launched Cortex M0 devices.