I have a sort of op-ed piece in Electronic Design today. Anybody who has been following my musing here (and, yes, I know I’ve not mused very much recently; must muse more ) won’t be surprised by anything I say.
The piece ended up being headlined “To Shake Its Malaise, EDA Must Look To Where Design Is Really Happening.” Journalists are constantly complaining about bad headlines being attached to their wonderful work, but in this case I think that the headline is a good summary of what I say.
The bottom line is that EDA, focused as it is on IC design in advanced processes, is focusing on a decreasingly important part of the overall electronic design process. Yes, you can’t design a leading-edge chip without EDA so the market isn’t going to go away. But most electronic systems use off-the-shelf chips rather than designing them from the ground up. There will always be a market for bespoke Saville Row tailoring of expensive suits, but the real market is at Macy’s, Nordstrom’s and Mens’ Wearhouse.
Here’s an example. The biometric company I work for has a fingerprint-protected USB drive product (that we got working the night before CES, it’s not just taping out a chip that comes down to the wire). It contains some flash memory, a USB and hardware-encryption chip (standard product) and a programmable Luminary chip (now part of Texas Instruments). The whole system requires a fingerprint sensor and an OLED too, which obviously can’t be integrated onto a custom chip in any case. Of course in volumes of hundreds of millions it would make sense to integrate the Luminary chip (which is an ARM processor with some standard peripherals) and the USB/encryption chip. But it will never ship in those volumes (I can dream) so I can’t imagine that would ever make sense. Although, as a long-term IC guy, it upsets my sense of elegance to have two chips that clearly “should” be integrated, it is simply cheaper to use two separate chips. Most electronic products are like this: a handful of highly-integrated but standard chips on a little circuit board.
One theme that runs through this blog is that semiconductor economics drives everything. Semiconductor is a mass-production process that can deliver very cheap chips but only if the “mass” in mass-production is large enough. Otherwise the fixed costs overwhelm: the cost of design, the cost of masks and the fab setup times. The only alternative is to aggregate end-user systems so that the same chip is used in multiple designs. FPGAs are obviously one form of aggregation, just buy raw gates and put them together later. The Luminary chip in the Biogy drive is another.
I certainly don’t claim to have all the answers as to what the big EDA companies should do. But somebody needs to be the Mens’ Wearhouse of EDA and serve the mainstream market, even though the unit price is lower. I guarantee it.