Two years ago I was asked to look at the assets of a failed startup. All its equipment and intellectual property were about to be sold for a pittance, so I was supposed to see what could be saved. The startup had found a way to build a cheap infrared camera, one that could see much deeper into the infrared than ordinary cameras. A typical silicon sensor can see into the near infrared (a wavelength of about 1000 nm), whereas this could see down into short wavelength infrared (SWIR, about 1600 nm). There’s a lot of this kind of light around even on dark nights – it’s emitted constantly by a layer of the stratosphere. They thought they could build a great surveillance camera, one that didn’t need any illumination at night.
It turned out that almost no one cared about that. If you put a surveillance cam outside, you also put a light there. The sort of places you really want to cover are truck yards, where it’s easy to steal a lot of stuff, and those always have lights. The people who really care about covering dark areas are ones who don’t want to advertise their presence. That’s basically military bases in hostile areas, and that’s not much a market.
So they went under. The technology was great – they had found a way to put a layer of germanium on a standard silicon sensor (germanium has a much lower bandgap than silicon, and so can sense lower-energy photons), so it was a simple add-on to existing tech. The people had come out of Bell Labs, and one of them was an IEEE Fellow, which is a real honor. I asked them what else you could do with such a sensor. “Well,” said the former CTO, “it turns out that SWIR light goes right through healthy teeth, but is absorbed by the water in cavities or other damage.”
I almost fell over. They had a way to do dental exams without X-rays, and they spent all their money on surveillance cams for Afghanistan. What a waste! “Wouldn’t that have been a better application?” I asked. “Our CEO and VCs were commited to the surveillance idea, so we never pursued it,” said the CTO. And now they can’t.
I’ve seen this happen lots of times. Someone builds something really nice that really works, but its basic purpose was wrong-headed. A few other examples:
- A wireless link capable of gigabits per second that’s meant to replace a $5 HDMI cable.
- A full custom chip for a proprietary disk interface (DSSI) that’s only going to be used on a few thousand workstations. That’s about $1000 of design expense per chip for a chip that only costs $20.
- A fully-programmable parallel image processor that spent all its cycles doing standard operations like video compression. That stuff should have been done in a standard block, leaving the processor to do novel operations that would distinguish the product.
- A chip that can do the latest and greatest video compression algorithm (H.264 SVC), one that’s so difficult that no one can actually decode it. Well, they could if they had the same custom hardware, but no one will buy that until video is available in this format, and no one will do that until everyone can decode it. Chicken, meet egg.
The engineers involved in all these projects did their jobs well. These projects were completed, and worked. It was the people at the top who screwed up. They asked for the wrong thing to be built.
That’s not always fatal. I once heard Wally Rhines speak about this. He was the manager of the group at Texas Instruments that built the first major digital signal processor (DSP) chip, the TMS32010, in the early 80s. He’s now the CEO of a large chip design softwarecompany called Mentor Graphics. DSPs are now a $10B product segment and are used in a vast range of applications, from cellphones to disk drives to audio. Even way back then TI knew they had a hot product on their hands. It was obvious what DSPs would be used for – speech synthesis. Everyone would love to have talking cars! TI had actually had a big success with the Speak & Spell, a toy that let you type a word and then hear it, and so knew that this was the wave of the future. Ten years later they looked back at the top 100 applications of DSPs, and speech didn’t even make the list. They guessed completely wrong, but the design was so flexible that it could be used for lots of other things. Few projects get to be so lucky!