Semiconductors: Where is the innovation?
You may have noticed that I have not been talking about marketing and media in EDA and semiconductors for a few months and I've been struggling with how to talk about it for some time. It's finally reach a head, though, so I might as well say it.
Innovation in the semiconductor sector is so rare now that it virtually does not exist. That may be why no one in the semiconductor industry, outside of the giants, does any marketing. But then, if you don't do marketing you won't know what needs to be innovated. Vicious circle, that.
I've had a love/hate relationship with EDA and semis for a long time. I've always thought they were incredibly important sectors and a source of great innovation. I've worked with some of the most innovative and groundbreaking companies in both sectors and I always hoped to be a part of it. But in the past few months I've come to understand what the VCs, entrepreneurs, government agencies and regional development agencies have been trying to tell me. There is no longer any there there.
In the past year I have seen three significant innovative technologies absolutely crushed because their market -- semiconductor design and manufacturing -- was just not willing to change how they do things. All they want to do is "reduce respins." That's it. Every VC I talk to has told me for the past three years that they know of no one in the industry that has a good idea anymore. What's more, they tell me, all the truly innovative people in the EDA industry have left it for completely different niches.
Last Wednesday I was attending a panel on worldwide innovation, sponsored by Silicon Valley Link and Eurocal Group. On the panel were executives from SAP and they were discussing how their company encourages innovation within and without. A lightbulb came on. these guys said their company doesn't set a budget for innovation. Their entire culture is based on innovation and as a result innovation happens. By contrast, EDA and semiconductor guys brag about the percentage of their budget dedicated to R&D and, as a result, they just make incremental advances on 20-year-old technology.
So what I have seen in the past few months is where all the innovation is coming from. And it is in software, not hardware. Apple has been successful in creating software-enabling platforms and when they could not find a semiconductor company to listen to them and make what they needed, they went out and bought a couple of them. Problem solved. Don't need EDA at all anymore.
Cadence is making noise about innovation through software with their EDA360 program. I wish them luck, but the rank and file of the industry doesn't get it. I know because I have talked to more than 60 EDA companies this year, desperately searching for some start-up that gets it and none of them do. i have found maybe three semiconductor companies that get it, but they can't get funding... because the VCs know that the tools they need don't exist. Even EDA Analyst Gary Smith says Cadence has a hard road in front of them because they lack the tools to make the vision happen, and no one is funding innovation in EDA so where will the tools com from?
So is this a bad thing? Not really. I've also been finding companies, that are getting funding, that are figuring out how to deliver functionality on off the shelf-hardware or letting their product reside on the cloud. The semiconductor and EDA worlds are quickly just becoming a convenience store to technology; a place where you drop in when you need an overpriced bottle of milk at the last minute.
So I have found the innovation centers I have desperately wanted to support. I will still look fondly back to my technology roots and will even take a meeting now and then, but the world has passed my old stomping grounds by. Time to move with it.
Hi Lou,
ReplyDeleteI have to strongly disagree on your "no innovation in semiconductors" conclusion.
It's easy to get jaded in an industry that's made exponential progress for the past 40 years and accept the exponent as status-quo, but the fact that we're now seeing 28nm and 22nm devices becoming reality - and that we continue to track reasonably close to Moore's Law is nothing short of a miracle of innovation.
This week, Xilinx and TSMC announced that they will be producing FPGAs with 2 million lookup tables! And, those devices are the first production of non-monolithic FPGAs - multiple known-good silicon slices assembled onto an interposer layer. A few weeks ago, Altera demonstrated FPGA transceivers running at over 20Gbps on 28nm fabric.
These are just a couple of the many major accomplishments we've seen in just the FPGA space - in just the past few weeks.
The big EDA companies are big because of their sales forces, not because they have a history of innovation. The majority (but certainly not all) of the innovation in EDA for ASIC/SoC design in the past 20 years has come from startups - and startup funding has become very scarce lately. In addition, with ASIC design starts declining, those big EDA companies are in the unenviable position of building increasingly complex tools for a vanishing audience. EDA for custom chip design like ASIC and SoC is on the verge of losing the economy-of-scale required to support a third-party industry. Tool innovation won't stop, it will just not necessarily be coming from EDA.
As a result, we've got companies like Cadence looking for opportunities outside hard-core custom chip design, Synopsys dabbling once again into the growth area of FPGA with their Synplicity tools, and Mentor continuing a strategy of diversification outside custom chip design that they've been on for more than a decade.
When we reach the point where nobody but companies like Apple, Xilinx, and Altera can afford to do custom chip design, EDA will need to have a different business model than it does today to survive. However, EDA's woes are not a reflection of no innovation in semiconductors. They're a reflection of the fact that big EDA has ridden the ASIC/SoC horse almost exclusively for far too long. And now, they're starting to pay the price.
Kevin, I have a lot of respect for your view of the FPGA industry. In fact, you might know it better than any other observer, however, note that I said that innovation is so rare it is virtually non-existant. That doesn't mean it is non-existant it's just not worth noting at the moment.
ReplyDeleteThe semiconductor semiconductor industry last year had sales of less than $220B, down from $240B the year before. FPGAs, according to Moshe Gavrielov, represent less than 2 percent of that. EDA sales are just a bump above that. But if you broke all of the innovation out of that market in the past 5 years it would represent nothing that would be considered even worth noting.
Both EDA and semiconductors have become service industries. With very little exception, the industry either does not know what the problem is that needs to be solved, or it really doesn't want to face it.
FPGAs are, I admit, a bright spot in the gloom, but when it comes to what matters to the rest of the world, it doesn't. What is happening is that the real innovators are using the FPGA platforms to launch their ideas. So the innovation is not coming our of the FPGA niche, but from it's customers.
That's my point. And I'm not seeing anything that tells me otherwise.
Lou, I'm in Kevin's corner. The industry innovates as much today as it ever has (maybe moreso because the physical and chemical challenges are so much greater).
ReplyDeleteAnd if the industry can't crack $300B a year in sales, well maybe that's huge success. It says that every year we make tons more products, while the price to buyers goes down. What's wrong with that?
Sure, maybe the DRAM industry has NEVER been effectively profitable ever, but faster denser memories are progress's pilings.
The iPhone doesn't happen without semiconductor innovation. Period.
Agreed, the iPhone doesn't happen without semiconductor innovation. Unfortunately, it's Apple that's driving that innovation, not a semiconductor company.
ReplyDeleteSit down and talk with anyone from the EDA industry and ask them, what's the most important problem you solve. 90 percent of them will say "reduce respins." So everyone in the industry is "reducing respins." Tell that to a VC and they walk out the door.
Sit down with anyone from the Semiconductor industry and ask them the same question and they will say, "smaller and faster chips." So everyone in the industry is making that happen... and the price to make them keeps going up. VC's walk away from that.
Talk to someone in the electronics system industry and they say "making more reliable products at a lower price with better profitability." The EDA guys and semi wonks say "huh?" so the systems guys go out and buy semi companies so they can get them to listen, or they hire software engineers to overcome the weaknesses of the hardware.
Maybe this is happening in big companies, and judging from the recent edition of EE Times Confidential that may be true. But there is not much happening in the places that have been known for innovation. And when it is, it gets buried.
I talked to a company earlier this year that taped out a product on $2M of investment with three engineers sitting in a single room. They used EDA tools from a company that is not the industry leader. But neither the EDA company nor the semi company have done anything to trumpet how they went about it. Why? Well if you let people know that you don't need $10M in EDA tools to make a successful, innovative product, you are going to cut into your sales.
Can the industry innovate? Absolutely. Do they want to? Not convinced.
Show me. I really want to know. Not in isolated incidents but show me a culture of innovation, somewhere, anywhere.
Maybe we should back up a step and find out who you're talking about when you say "semiconductor companies" - I think of Intel, TSMC, AMD, UMC, Toshiba, Fujitsu, IBM...
ReplyDeleteI see plenty of innovation there. Granted it just gets us to the next process node once every two years - like it has for the past forty. Still, we haven't missed a node yet, and that's truly remarkable!
My examples were related to FPGA companies, but they certainly aren't just FPGA innovations. 28nm and 22nm semiconductor processes are monumental innovations - and they go far beyond FPGAs.
Thinking of a new process node as "smaller and faster" really hides the point. More realistically - they're "bigger and lower power". A new process does exactly what you're asking for - it gives us the ability to make more complex systems with higher reliability for less money - while consuming less power and occupying smaller form-factors.
As we move forward, re-spins do, in fact, become the biggest problem. It is so expensive to make a custom device, that an extra re-spin or two can put you out of business. With that kind of fear-factor hanging over their heads, customers of EDA respond better to the "reduce respins" message than the "improve productivity" or "shorten design cycles" messages. This is primarily messaging, however. Behind the scenes, EDA has to work really hard just to do the same thing they did before - only on the newer, more complicated process with designs that - once again, just doubled in size.
It doesn't sound like "innovation" when you could buy place-and-route software twenty-five years ago, and you can still buy place-and-route software today, and both tools took a day or more to complete a full-chip run.
But, in order to have that full-chip run succeed on a multi-billion transistor design at a 28nm process pitch with eight times the number of metal layers, far more complex design rules, and about 10,000 times the number of objects to be placed-and-routed - there is some serious innovation required. That's not about "reduce re-spins" but it's also not making big news or exciting press releases. What's more - with the dramatic reduction in design starts, there are probably about 10x fewer customers available for the new place-and-route software than there were for the 25-years-ago version. That makes the business side of EDA a challenge.
The "culture of innovation" you're asking for is self-evident, proven by the fact that we now have devices being manufactured at these stunningly small process nodes.
I think we have a different definition of innovation. See tomorrow's post.
ReplyDelete