History suggests that a company with two 8/16-bit microcontroller architectures and at least two different 32-bit CPU architectures will not survive very long. One of three outcomes is likely: the company will crash and burn in the effort, one or the other architecture will go into the dustbin of history, or the company will sell off one of them, the way Microchip plans to sell Atmel's nonvolatile memory and RF and automotive businesses to On Semiconductor.

A look at the industry's history reveals that trying to support more than one architecture just does not work. In the recent past, Intel Corp. was juggling three different architectures, targeting three different markets: the DEC StrongARM-derived PXA2xx targeting mobile and consumer designs, the IXP4xx family of network processors it also acquired from DEC, and the X86 architecture targeted at desktops, laptops and servers. What we have now is a single architecture from Intel, targeting all these segments.

Further back there is Advanced Micro Devices, which in the 1990s had to make a choice between supporting its X86 strategy in desktops and servers and its internally developed 29000 RISC architecture, which at the time had a strangle-hold on the high-end printer market.

More recently AMD tried to juggle two different efforts, its X86 in desktops and servers and the MIPS-based architecture acquired from Alchemy for mobile and consumer devices. It is now focused on a single X86 strategy.

In the 32-bit arena Atmel will have to make a choice between two different RISC architectures, the MIPS-based one it uses in its new PIC32 and the ARM-core based 32-bit family that Atmel has proliferated throughout the embedded market.

The MIPS architecture is the one EE and CS students read about in their text books. On the other hand, the ARM RISC is widely supported and licensed to virtually every serious player in the market. And while the MIPS architecture has an impressive repertoire of development tool support, it can't compare with the support available for the ARM. It will be a tough choice, but at some point Microchip will have to make a decision as to which to support totally.

Microchip faces the same problem in the 8/16 bit microcontroller business. Despite the fact that Microchip is No. 3 in the MCU market with its PIC family, Atmel's AVR family is a well-respected RISC MCU architecture, especially in Europe and in certain market segments such as automotive and industrial. There is no way the two architectures will coexist within the same company.

History again tells us that it just doesn't work. Intel, when it was a serious player in the microcontroller business, had two separate architectures, the 8/16-bit 8048/8051, now discontinued by the company but second-sourced by eight to ten other manufacturers, and the 16-bit 80196, which now survives only within the company as a specialized CAN communications controller.

The same story is repeated at Motorola and STMicroelectronics. Both at one point had multiple MCU families and each now supports only one.

With this history as a guide, if Microchip's acquisition of Atmel is successful, Microchip will have to make choices about which architectures to support fully and which to either let die for lack of support or sell off to interested buyers.