Blog Post

Startups Abandon Moore's Law

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

Producing a cutting-edge chip today is more difficult and more expensive than it’s ever been. With a foundation in Moore’s Law, which dictates that the number of transistors on a chip doubles every two years, the rules of economics and physics are stretched. Semiconductor makers are spending billions, to either build joint manufacturing facilities (called fabs) or to create research agreements that help offset the basic costs and challenges involved in trying to push a technology forward.

With costs for the latest technology rising, smaller startups and their venture backers have taken a look at the benefits of keeping up with the Intels, and decided that in some cases, old-school technology may be the best way to make a product — and a buck.

Even as costs have risen on the manufacturing side, the exit opportunities have shrunk. In 2007, which was a banner year for IPOs, nine chip companies made it to the public markets — and all but two of them are currently trading at below their offering-price levels. Also last year, 12 chips companies were purchased, according to data from the National Venture Capital Association, while back in 2006, just two chip companies went public, while 17 were acquired. Given the state of the exit markets, it’s no wonder that venture investors are scrutinizing the returns on the $9.91 billion they’ve invested in the space since 2003.

But those that are sticking with chips are searching high and low for capital-efficient chip startups. Jim Jones, a general partner at Scale Venture Partners, says he doesn’t want to put more than $50 million into a chip startup to get it to profitability, because beyond that point it just doesn’t generate the returns needed to justify the investment.

So in an effort to save money and preserve their returns, Jones and other VCs are investing in the next new thing — built using technology from the previous decade. Instead of following in the footsteps of the top chipmakers such as Intel, AMD and Samsung, which are packing more onto smaller chips, startups are turning to process technology from up to 10 years ago.

Jones points to Auvitek International, a startup designing chips at the 130 nanometer node — the node at which Intel was designing products back in 2001. The company, which makes a modulator that receives over-the-air-broadcasts on TVs or PCs, has managed to get its products to general availability with $13 million in funding. Another Scale portfolio company, Xceive, is making an analog tuner at 180 nanometers; it’s been selling chips for two years on $25 million in capital.

Many consumer chips targeted at consumer applications, like those above, don’t have to be that fast. Other types of chips that don’t require speed as a differentiator, such as power management chips or RF chips that have antenna modifications, are also viable targets for older manufacturing processes.

Drew Lanza, a general partner at Morgenthaler Ventures, says the firm realized about two years ago that the current race to the leading edge wasn’t going to mean viable exits, so he started to look at investments beyond those in processors and graphics-oriented chips. So far Morgenthaler has invested in SiPort, a high-definition radio chip maker manufacturing at the 130 nanometer node, and ZeroG, a low-power 802.11n chip.

“I think we’re seeing a fundamental shift here,” Lanza told me. “For a long time, startup companies helped drive the leading edge of Moore’s Law, and I don’t think we’re going to be doing that anymore. We may have EDA startups and we may back manufacturing equipment, but I think we’re off Moore’s Law, probably forever. We used to be maybe one generation behind the leaders, but we’re three to four generations behind Intel now.”

This trend may make sense for startups in certain sectors, but when it comes to industries such as communications or processors, the need for speed at lower power means that startups must raise enough money to build at 65 nanometers. That can be pricey; it costs about $30 million to design a chip at 90 nanometers, but those costs can reach $100 million for a startup seeking to design at 65 nanometers.

In other cases, it makes sense to start out on a larger node, gain market share that can help justify the costs of moving down a node, and then scale the design down. Ubicom, a 12-year-old startup that has hit its stride with consumer-targeted network access chips, is currently making the jump down to 90 nanometers from 130 nanometers. Such a shift can cost up to $10 million, but the increased speed can be worth it. Startups may be abandoning Moore’s Law, but they’re still going for speed — just more slowly.

14 Responses to “Startups Abandon Moore's Law”

  1. Two things:

    1. Analog doesn’t scale. Digital does. The startups you’ve mentioned using older technologies are all mixed-signal analog names.
    2. You need smaller process nodes to also be cost-efficient in the semiconductor space. Cramming more chips on a single wafer amortizes the startup cost on a larger number, and lowers your COGS.

    If you’re a startup proving a new design, you can do a proof-of-concept in an older node, and port it once you get funding on the basis of the earlier design.

  2. Another start-up of note that jumped off the Moore’s Law treadmill is Luminary Micro. It’s building very low cost ARM microcontrollers using at least N-4 process technology. The chips are small enough that they don’t need cutting edge process and the older process is really cheap.

    Some start-ups still do compete on the leading edge process, like Raza Micro, P.A. Semi, and Cavium, but the costs of funding those start-ups is very high.

  3. For processors, the tech press and analysts all focus on Moore…blah blah double. Blah blah Physics. But if you speak to the new, young minds that are coming up in todays academic research kitchens, the new frontier might not be gate density or Moore at all – it’s the architecture.

    The CPU industry has been captive to the register, program counter, load/store architecture since…well, the beginning. Any innovations that came later, as alternatives (some actually based in very old ideas), they were swept aside by the juggernaut that was the silicon boom of the 89’s and 90’s.

    Liberating computing from these legacy load/store architectures will require small innovative teams to create new, non-register based systems and the software and systems architecture to match.

    Then we shall see how Moore predominates in a world where the sheer number of gates is the controlling performance factor.

    Quantum? Maybe, but I see a more conventional, bipolar or MOS based future, just with a new type of program operand and data model.