Call it aggressive; call it innovative; call it bat-poop crazy. Whatever your views, Tabula certainly had interesting ideas about how to design an FPGA. But now it’s gone.
The Santa Clara–based startup will close its doors at the end of next month. The staff have all been given their walking papers. The furniture and fittings will presumably go up for auction, a common-enough sight here in Silicon Valley. And the technology? Tabula’s secret sauce may wind up in the hands of an investor, or a new startup, or some big established semiconductor vendor, or a patent troll. Too early to tell.
It all looked so promising. We covered Tabula early on here, and more recently here. When the company came out of stealth mode in 2010 and secured a whopping $108 million in financing, the Wall Street Journal called it “the largest single investment in a semiconductor company in the last 10 years – possibly ever.” Take a moment to brood on that point: The largest investment in a semiconductor company since 2000. That, my friends, is a big bet. One of those investors characterized it as “a swing-for-the-fences kind of opportunity.”
It didn’t hurt that Tablula was going to fab its chips at Intel. Intel! In a foundry agreement with a startup! That’s like discovering that your high school chess champion is dating an international supermodel. Props, dude. Tabula’s reputation in the nerd community rose dramatically. The company had an all-star cast of certifiable geniuses, a boatload of VC money, and a lot of years of research and development under its belt. And it had something cool called SpaceTime.
You see, Tabula had created a 4-dimensional FPGA. Not only did it encapsulate logic gates like a normal FPGA, it also conceptually (though not physically) stacked layers of these circuits vertically (the 3rd dimension) and then switched them rapidly over time (the 4th dimension). It was revolutionary! It was awesome! It was insanely great and promised to transmogrify everything we knew about the FPGA and ASIC markets.
But like so many of us, it was too cool.
I’m no FPGA expert – you can read Kevin’s analysis for that – but I think that Tabula failed because its sex:food ratio was too high. It was too exciting, too innovative, and too strange. In short, it was too sexy. And while that’s exciting to dream about, it’s not the choice you make when you’re in your cubicle planning your next project. We might fantasize about someday buying that new sailboat or a flashy new car, but day to day, we eat cafeteria food and drive around in a rusty Honda Civic. It’s nice to have aspirational goals, but the basic and the familiar are what pay the bills. There’s a reason fast-food chains are more profitable than boutique restaurants serving probiotic locally sourced free-range vegan fare.
It’s a trap that a lot of engineering firms fall into. They’re too aggressive; too enamored with their own cleverness. They forget that they’re selling this stuff to people. And people make irrational decisions. Or more accurately, they make decisions that you think are irrational but that make perfect sense from the buyer’s point of view.
Let’s review a few of the inventive technologies and products that have failed over the years. It’s easy to find examples. There’s Intel’s i860 microprocessor, Freescale’s 88000 CPU family, the Transputer, Intel i960, TI 32000, AMD 29K, and many more. Plus any number of hardware-design languages, such as C-to-Verilog, CoWareC, ELLA, Lava, and many others. Or programming languages like Occam, Limbo, FortranM, Fortress, ParaSail, and innumerable other graduate-student projects. Remember Multiwire boards? The list goes on and on.
What these failed technologies have in common is that they were too hard to learn and use. Yes, the microprocessors, the programming languages, and the HDLs were all very advanced and well thought-out. But to their potential customers, they were just… strange. They required a new way of writing code, or a new way of doing board layout, or a new set of development tools, or even a new team of developers. Customers had to sacrifice their experience and familiarity in exchange for promised new features and benefits. And for most, that wasn’t a good trade.
That doesn’t mean we can never innovate – far from it. But it means we have to remember that somebody, somewhere has to buy and use our products. And it’s that last step in the chain that is so often forgotten.
Remember that nobody – nobody at all – cares about your new product as much as you do. Not your spouse, not your boss, not your investors, and certainly not your customers. After all, you’re the one putting in the long hours in the development lab, not them. You’re the one who started the company, or created the technology, or spearheaded this new project. If anyone else was a gung ho as you are, they’d be working right there next to you.
Stated another way, all of your vendors, suppliers, and customers care less about your product than you do. They don’t see the big benefit and they probably don’t care very much. They’ve got their own projects and products to worry about. So as awesome as your new development undoubtedly is, few others see it that way.
More specifically, they’re not going to get excited about the implementation details. (Again, if they did feel that way, they’d be working alongside you.) You see the inside of it; they see the outside. Only you know how clever you had to be to get it working just right. All they see is a new box, a new chip, or a new design problem that they’re going to have to wrap their head around. Imagine if someone designed a new car that got 1000 MPG by using gopher sweat as fuel. That’s a remarkable piece of engineering, but… where am I going to fill my tank with gopher sweat? It’s the same problem faced by Nissan, Tesla, and other electric-car makers: outstanding engineering held up by niggling customer resistance.
In the perfume industry (which is about as far removed from electrical engineering as you can get), the dominant cost of doing business is buying space on the retail store shelves. Department stores don’t stock perfumes just for fun; the perfume “developers” have to pay for placement, and the closer you get to the front door, the more expensive the subsidy. The perfume ingredients and manufacturing are negligible costs; it’s getting up the customer’s nose that’s expensive.
The same principle applies to software, systems, and devices. All the elaborate engineering in the world won’t sway a customer unless you can actually sway the customer. It’s hard – very hard – to see your new baby from a customer’s point of view and to understand what they’ll like and what they won’t like. That’s why grown-up companies hire marketing people, and why those people often don’t have technical backgrounds. That’s not a mistake; that’s deliberate. You want somebody flogging your product who isn’t emotionally tied to its development or wowed by the awesome cleverness that went into it. You want someone with the customers’ mindset.
Tabula was a victim of its own cleverness. It was undoubtedly a triumphal piece of engineering. The hardware was exotic; the software was complex; the tools were advanced. But those characteristics made it intimidating for the plebes who had to actually use the devices. Dialing back the wow factor might have kept the company afloat a bit longer, so the rest of us could catch up to it.
I spent too much of my marketing life actually believing it when prospects said they were interested in a new product idea. When it came time to pay for it, they never were. So, at the risk of sounding like a soulless money-grubber, I came around to realizing there’s only one way to be sure a prospect really likes the idea: “Show me the money.”
I think you missed a huge reason companies don’t buy into products from a company like Tabula. Engineers can’t bet the farm (ie their project and thus perhaps their career) on a company that has a huge probability of not being around when production starts.
A while ago we asked our boss if we could use a new wizz-bang device from a young company. The answer was “Yes, but you should consider it a career threatening decision.” We went ahead and dropped in a Xilinx 2000 series FPGA. Anyone who designed in Tabula ain’t so lucky!
As far as I know, the Tabula devices never stacked “layers of circuits vertically”. I think that the author misunderstood the use of the Intel “22nm process with 3D Tri-Gate transistors” which is more commonly referred to now as 22nm FinFET.
There were a lot of good engineers at Tabula many of which got their start at Xilinx or Altera, but the basic architecture premise that rapid reconfiguration, aka SpaceTime, would provide a significant cost, performance or power benefit to the end user was flawed in my opinion.
No, I totally get that Tablula’s chips didn’t physically stack their circuits, like some 3D chip techniques. I had hoped the article made that clear. And I wasn’t referring to FinFET, either.
Rather, the “3rd dimension” was the appearance that there were multiple circuits one atop the other. Even though they weren’t physically implemented that way, that’s how Tablula described it on more than one occasion. In concept, you had alternative circuits overlaying each other, switched in time. The reality, as usual, was a lot messier.
Another one bites the dust. Wow, 200+ million down the toilet. That has to be one of the most spectacular semi startup fails ever.
Actually I think it will leave a smelly stink on the semi startup cause. If you can’t succeed with 200 Million, why bother?
The VCs should have pulled the plug years ago when the original one was basically a space heater due to the incredibly high power.
They had major power and tools troubles I doubt were ever really solved.
The FPGA market is a duopoly. Xilinx and Altera have the solid lock.