10 Comments

Summary:

The electronics inside today’s hottest connected devices are often custom and difficult to engineer because the big chip firms don’t offer the information entrepreneurs need. How can the industry fix this?

Making this board was unduly complicated.
photo: Scanadu

We are awash in new hardware products that promise to make us healthier, connect our homes and even change how we play. But amid this bounty of connected devices there is a little discussed opportunity for a chip firm or a startup — making it much, much easier for people to build out small, smart boards for prototyping and finished products.

Inside every one of your connected devices is a small green or red circuit board that the various sensors, radios and processors are soldered onto. In the prototyping stage these boards might be a Raspberry Pi or an Arduino combined with other boards (shields), but when it comes time to build a real product in units of 1,000 or so things become … difficult. Speaking at SXSW interactive event in Austin Daniel Conrad, the founder of Beep, expressed frustration with buying chips for his in-room music streaming product.

The Beep device.

The Beep device.

He explained that unless engineers are looking for 10,000 chips it’s tough to get a foot in the door. His search for a small embeddable computer that ran Linux, but was cheaper than a Pi led him to a variety of chipmakers before finally being pointed to the brains inside a router. Routers require processing power and also run Linux, so Conrad tried it.

However, routers aren’t engineered to connect intermittently to a USB port, so the contact was iffy and the resulting sound quality for his device was awful. He found himself searching Alibaba for chips and found an alternative. It came with its own problems. He has since solved his issue after trying several other vendors but it took time, money and a bit of luck.

He’s not the only one. As far back as a year ago I spoke with David Merrill, the president and co-founder of toy company of Sifteo (see disclosure), who said one of the biggest challenges and reason for his product’s high cost was the lack of modular components and easily integrated boards. Each Sifteo cube has a custom-based board, and getting them designed and built was a source of frustration.

Wireless Sifteo cube game

That same frustration was expressed on Monday in a conversation with Walter De Brouwer, the founder and CEO of Scanadu, whose company has built a palm-sized medical scanner that measures blood pressure, heart rate, respiration and other vital signs. The Scanadu device is shipping at the end of the month to its initial backers and will retail for $199, a price De Brouwer says could be cut significantly when they can get all of the components on the Scanadu circuit board (pictured at top) on a single chip.

The challenges each of these entrepreneurs face in building their hardware could be addressed with time and money, but if the hardware renaissance is to continue perhaps we need a few more players in the ecosystem to help solve this problem. There’s the challenge of finding the right chips for the job, offering the design expertise required to place components ranging from sensors and microcontrollers to FCC-approved radios in the best and most compact position, and then manufacture them at a smaller scale than the large companies in China. Faster integration of multiple components into one system will also be a boon.

I’ve seen such small-scale manufacturing operations in the U.S., such as on the floor of the Modular Robotics manufacturing site, where a pick and place machine that was at one time used to make bombs now manufacturers toys, but I’m not sure how to get the chip firms to rethink how they sell a categorize their wares.

Modular Robotics CEO Eric Schweikardt standing next to the pick and place machine.

Modular Robotics CEO Eric Schweikardt standing next to the pick and place machine.

Large companies from Intel to Texas Instruments have got on board with boards aimed at the maker movement, hoping to get developers designing products with their chips, but many of these devices are not made for building out actual products in units of 500 to 1,000. Maybe there’s a role that a company like CircuitHub or SparkFun could take on. Maybe we need a new startup that can open up the online equivalent of a RadioShack with higher-level processors and components.

There’s a pain point here. Who will solve it?

Disclosure: Sifteo is backed by True Ventures, a venture capital firm that is an investor in the parent company of Gigaom.

  1. On the chip side there are things like the PSoC from Cypress.
    On the manufacturing side 3D printing should help.
    Also 3D printing related, if it takes off for consumers, there will be a lot more options for everybody since someone has to provide the electronics and services to enable users to make things at home.
    Anyway , enabling individual users to design and make things has been a topic for a while so maybe some are exploring how to get there . Could be, for example, a followup to project ARA for Google.

    Share
  2. The problem is caused by a combination of things, but there are two key culprits: the complexity of SoCs. and the business models of most SoC vendors.

    The hardware reference manual for even a $5 SoC can run over 1000 pages, and even they don’t provide enough details for many applications. As a result, hardware and embedded software engineers need support resources at the SoC vendor, and those vendors don’t really want to provide that kind of support for a company that might buy only 50K $5 or $10 chips a year. So they dump that responsibility off to one of the two big distributors (Arrow and Avnet), who just don’t have the resources to adequately support the hundreds of companies they carry.

    Companies like TI and FreeScale address this problem by open sourcing as much software as possible, and making documentation and reference designs readily available. They also encourage users to be actively involved in forums where problems can be resolved by other users, or their employees can step in and answer questions when appropriate. However, most of the other SoC vendors don’t follow this model, and keep everything under NDA. Which means only large customers can get access to it. It’s a ridiculously paranoid policy, but unfortunately one that has been adopted by many SoC vendors.

    Also unfortunately, TI and FS don’t make very many inexpensive SoCs, so it’s tough to build a consumer product based on one of their chips that will cost less than $50. There are several chip vendors in Asia that do make those chips, but they also cannot support engineers in the U.S. easily, there are documentation issues, and they have little or no distribution here. I have opened up many products to find out how they could be built so inexpensively, only to find a chip that cannot be sourced here, let alone designed in.

    There is a great opportunity now for hardware engineers to create new companies in the U.S., but it’s not going to happen as widely as it could if the chip vendors don’t’ make it easier for engineers to design in their products. I don’t see TI and FS SoCs getting cloned often, so I don’t know what they are afraid of.

    Share
    1. Excellent insightful overview. You nailed all SoC issues. Note besides IP theft, there is fear that info about a defect will be magnified & exploited by a competitor to steal business from big accounts.

      Share
    2. The IP issue is real, as companies must safeguard against any innovative circuit blocks (or even robust, reliable infrastructure blocks) from being purloined by competitors. As the Snowden files and its repercussions have revealed, industrial espionage goes in all directions.

      But the big issue when it comes to custom silicon is the gulf between your typical startup and Big Silicon companies like TI, Freescale, etc. with armies of highly educated and highly skilled engineers at high levels of compensation. Fabless silicon startups have fallen out of favor with investors, save for a few exceptions like Calxeda (which shut down at the end of 2013) and Adapteva (a going concern, bootstrapped to start, with their first reference board out). A couple of niche FPGA companies (e.g., Silicon Blue) have graduated from startup to SME but their market shares are miniscule against the Big Two (Xilinx, Altera) and the Next Three (Actel Microsemi, Lattice, QuickLogic)

      Share
  3. There’s also the FPGA vendors that can help. Xilinx has the Zync series offering dual-core Cortex-A9 CPUs, along with standard SoC components and interfaces and a programmable logic section for customization:
    http://www.xilinx.com/products/silicon-devices/soc/zynq-7000/index.htm

    I think the problem is that these hardware startups are not comfortable with customizing chip designs.

    Share
    1. Those Xilinx Zync parts are $50 each in large qty, for the cheapest versions. The design tools are not cheap, either. It’s not a part that lends itself to a start-up that focuses on consumer products, or doesn’t have significant venture funding lined up.

      Share
      1. While the Zynq (silly Xilinx spelling) is a bit pricey, it could be used for a Gen 1 version and then turned into an ASIC for a less expensive Gen 2 product if Gen 1 gets traction. A lot of successful consumer product sell for more than $100.

        As an example, Technical Illusion kickstarter for their CastAR augmented reality glasses were prototyped with discrete components and an FPGA. The final product will have an ASIC.
        More details here: http://technicalillusions.com/wiki/home/

        Share
        1. As I said, this is not a product for anything but a well (VC) funded start-up. When you are talking about ASICs, it’s usually beyond the reach of small startups. And if you have raised lots of funding, then the problems faced by small startups don’t apply – the capital will attract the vendors.

          A $50 FPGA will not yield a $100 consumer product. It will take volumes in the hundreds of thousands to justify designing an ASIC.

          Share
          1. I second Ken’s comments, as Big Silicon companies like TI would only engage in designs that promise profitability in volumes of hundreds of thousands. TI has publicly acknowledged (in a monthly newsletter from Synopsys, the largest EDA company) that dud first silicon for an OMAP chip (sophisticated mobile SoC) would cost them $2 Million. And TI has abandoned mobile chips because Apple and Samsung have gone vertical and pulled the bulk of that market in-house. The consequences included the shuttering of a major R&D center in Nice, France.

            Share
    2. Thanks, Kevin, you beat me to this. There is a big gap between where electronics hardware startups are, most are designing at board level with standard components, versus SoC companies on the scale of (e.g.) Qualcomm or Intel.

      Yes, Xilinx’s Zynq (that’s a Q at the end) provides a programmable silicon platform with an ARM dual core in the middle and some standard interfaces at the boundary. Altera has a similar one, too. In addition, for design creation, I would recommend that hardware startups look into ESL (electronic system level) design with C/C++/SystemC rather than RTL with VHDL or Verilog. As it implies, C-based design can be more system-level rather than logic block level, and C knowledge is widespread so there is no need to learn another language.

      As a disclaimer, I am with an electronics design software startup called Space Codesign but I will only say that the Resources section of our web site (I hope that the address should be obvious!) has some helpful information for those who are curious about getting to the next level, in being able to direct the silicon as well as the software in their product design.

      Share

Comments have been disabled for this post