“Gallium Arsenide (GaAs) is the silicon of the future. Is and always will be.” That was the humorous perspective circulating among analog technologists in the late 1980s. In a similar vein, I can't count the number of times I've had people ask me when will the world see a field programmable analog array (FPAA) — the analog counterpart to the digital field programmable gate array (FPGA).
That is also the future of analog. And just like GaAs, it will remain always in the future for one simple reason: Analog is physical whereas digital is algorithmic. Unless someone can 3D-print an analog IC on your desk, FPAAs will never come to pass.
That's right. I used the word “never.” And yes, I know there are things like pSoCs from Cypress Semiconductor. But those products are not FPAAs. Rather they are microcontrollers with digital and some reconfigurable analog thrown in.
It isn't that people haven't tried to do an FPAA. In fact, the world is littered with the remnants of FPAA dreams gone awry and, along with them, about $1 billion in risk capital down the toilet. You would think that after nearly 30 years of trying, someone would have figured it out by now if it were possible.
A while ago I looked into the history of FPAA technology and compiled this timeline:
- 1988 California Institute of Technology, US: programmable fabric of current mode devices
- 1990 Pilkington Microelectronics, UK: programmable switched capacitor arrays connected around amplifiers
- 1994 International Microelectronic Products, US (European IMP team reincarnated today as Dialog Semiconductor): EPAC electronically programmable, predefined blocks
- 1996 Zetex Semiconductors Ltd., UK: Electronically programmable, predefined blocks
- 1997 Faura, Spain: Programmable, predefined blocks with a huge digital core
- 1997 Motorola, US: Bought rights to Pilkington (see above)
- 1998 Motorola and IMP withdraw from the FPAA market
- 1999 SIDSA, Spain: FIPSOC similar to Faura above
- 1999 Lattice, US: ispPAC, programmable predefined blocks
It was in 1999 that EE Times contributor Stephan Ohr summed up the problem in a piece he affectionately titled “Programmable Analog-here we go again.” Ohr concluded that all of the pre-2000 techniques suffered from the idea that customers would put a $7 to $20 part in a $1.58 socket.
In 2002, Cypress Semiconductor introduced the pSoC for about $1.58. Nice try, but no FPAA. The last attempt I am familiar with was GTronix in 2002. GTronix, or more likely its investors, gave up in 2010. While Lattice's ispPAC is still around, barely, and pSoCs seem to enjoy some success in reconfigurable analog, the rest are history.
Of all the players I listed above, I feel compelled to hand out the FPAA Courage Award to the team from Pilkington. Pilkington's technology started with its roots in the UK in 1990. Motorola, in 1997, and despite my uncompensated recommendation that it pass, acquired the technology with dreams of selling FPAAs at FPGA prices.
Motorola gave up shortly thereafter, and the technology moved back to the UK under the name of a company called Anadyne Microelectronics plc. It changed its name in 2000 to the much catchier Anadigm Inc. and established an office in Silicon Valley. Twenty-three years after Pilkington's start, Anadigm still has a website, but its address on Google Maps suggests a mailbox in Arizona. So I'm not precisely sure of its status, although Reid Wender listed Anadigm as one of his company’s competitors, so who knows? Maybe someone can update me on its status so I can hand-deliver the plaque.
Look, FPAAs are a non-starter. And I don't say this as just an outside observer. My bones can also be found in the scrap heap of FPAA dreams. But if there is a viable programmable analog technology, it will be found in the form of an analog front-end IC filled with multiple data converters and mated against an FPGA. Then non-analog types can GUI an analog system very quickly without worrying about 15mV op amp offset voltages.