Al Wegener and I first spoke back in 2004. Al is the founder and chief technology officer of Samplify Systems. Over the years since we first met, I’ve reported on Samplify and its evolution (electronicdesign.com/article/analog-and-mixed-signal/Analog-Mixed-Signal-Companies-Take-Many-Paths-To-Success.aspx). Since Samplify is a successful startup in the analog chip business, and since Al is still one of its key executives, it seemed to be a good place to start this series on the Mind of the Entrepreneur.
DT: As I recall, when we first met, you wanted to tell me about lossy and lossless compression. You had this idea about IP for data conversion. The basic concept was that, if you knew something about the nature of the signal to be compressed -- if it wasn’t totally random, it was possible to apply a certain type of lossy compression algorithm to it and still achieve lossless compression.
At least, that was the idea I believe you started with. Today, you have a company that seems to be deep into the ultrasound hardware business. How did the evolution happen and what’s going to happen next?
AW: I started thinking about this business in September 1999. It came about because I worked in the pro-audio business in the mid-1990s, and I kind of fell in love with audio compression technology. I left the audio business in 1995, but I continued my interest in lossless compression of audio signals. I was just intellectually curious about the compression technology.
Then, in 1999, a friend in defense electronics and I talked about capturing radar signals on disk drives, which at that time were still quite expensive. My friend felt that if they could get 2 to 1 compression of radar signals, using my audio compression algorithm, it would be a big win for his company.
At that point, yes, the/my business model would have been built around licensing IP technology. That was an eye-opener. When I first applied the technology to his radar signal it made them 10% larger. It took me a couple of weeks to figure out why my audio compression algorithm did not effectively compress radar signals Once I figured out how to modify my audio compression algorithm to work on radar signals, in September 1999, I incorporated Samplify as a Nevada LLC. I incorporated in Nevada because it only cost $85.00; in California, it would have cost ten times that. From 1999 to 2006 I was basically just fooling around with algorithms for compressing signals from data converters. I- didn’t have a specific market in mind for the product.
EVOLUTION OF A CORPORATE STRATEGY
In 2003, a neighbor of mine who was an investment banker, said, “this is an interesting idea. Who would use it?”
I said, “I’m not sure. I suppose that data-converter companies could use it. I hear that medical imaging involves processing a lot of data. Computer tomography, scanners, base stations . . . . I had a lot of possibilities, but I didn’t really know if those potential markets actually had bandwidth problems.
In 2006, when I was working for Texas Instruments, I approached them about putting compression into data converters. I had a carve-out in my TI employment agreement for my Samplify technology, since I had developed that before I joined them. TI declined the opportunity to use compression in data converters. But after that I couldn’t let go of the idea.
Continue on next page
So I contacted Bruce Sachs, who is the managing partner at Charles River Ventures, which became one of ’Samplify's first VC investors. Bruce and I had gone to college together. In March 2006 I called Bruce and said, “I’ve got this idea, I can’t get it out of my head, and I would like your help in sorting it out whether this might be an idea that your VC firm would fund?”
Three months later, he had seen enough from me, and heard enough from potential customers, that he said, “OK, I’m going to give you a little money to get started. I want you to quit your TI job and see if this works.” I left TI in July, 2006. In September, I hired Tom Sparkman as CEO in October 2006 and, together, we raised our first round of funding, $6.5 million.
Then we hired some employees and started working on an A-to-D converter with built-in compression.
CROSSING THE SOFTWARE/HARDWARE DIVIDE
DT: So at that point, you had already moved away from the idea of simply licensing the compression technology to companies that made data converters?
AW: Originally we did think we would license the technology -- but as IP netlists for Xilinx and Altera FPGAs, not as an IP block internal to ADCs. High-speed data converters need to deliver their bits to some other device, and typically, that’s an FPGA.
But the reason for turning to a business model based on hardware is that few IP companies are adequately rewarded for their IP. The semiconductor industry includes some notable exceptions, like ARM, Qualcomm, and Rambus. In contrast, if a company puts its IP into a chip, customers will pay you for it. This is the advice that that I got from Geoff Tate, the former CEO of RAMBUS. Geoff and I had several discussions while I was trying to decide whether to leave TI are not, and he counseled me, “Al, wrap your IP in silicon.”
In essence, everybody in the data converter market that we were addressing gets really good profit margins, typically above 80%. So we figured that making A/D converters with integrated compression was the proper way to monetize our IP. We would let compression be the differentiator that makes our data converters better than those from other suppliers.
THE PROOF IS IN THE HARDWARE
DT: That means dealing with foundries, how did that work out?
AW: It worked out very well. Our first chip, the SAM1600, is a 16-channel, 50-Msamples/s (per channel), 12 bits per sample converter, targeted for medical ultrasound. It is still the lowest power, highest channel count ADC in that market.
To accomplish that design, we partnered with another company. We designed the compression piece. The company we worked with had a lot of background in that kind of ADC and they had worked with UMC at the 130 nanometer process node. The only thing different with our chip compared to what they’d done before was that we wanted more channels. The only parameter that was a challenge for this design house was those extra channels.
COMPRESSING CUSTOMERS’ TIME-TO-MARKET
DT: From our previous conversations I know that one thing that made life more complicated after that was that you suddenly had to know more about ultrasound beamforming.
Continue on next page
AW: That didn’t happen at precisely that point. Something else came first. Imagine you’re a startup in Silicon Valley, and had built an ADC, with compression, that you want to start selling to the medical-instrumentation companies. We quickly found out that the demand for ADCs in that market was in China.
So we go over to China, and there, the engineering talent is used to working at a different integration level than U. S. engineers. - end. Because ultrasound front ends must deliver more than 100 dB of dynamic range, the PC boards that hold the low-noise amps (LNAs), variable-gain amplifiers (VGAs), and A/D converters must be carefully designed and integrated. By offering Chinese engineers an pre-integrated ultrasound front end (AFE) subsystem (where Samplify had already integrated the LNA/VGA with the SAM1600 A/D converter), the Chinese engineers could validate the subsystem much more quickly than they could have designed it themselves, so they get to market faster with Samplify’s AFE, rather than just with our SAM1600 A/D converter.
To do that, we partnered with Maxim Integrated Products to do the low noise amps and the variable gain amp in an integrated solution: two of Samplify’s 16-channel A/Ds paired with four of Maxim’s 8-channel LNA/VGAs.
Okay, after that is where the beamforming comes in. That’s another function for the FPGA. At Samplify, we already understood all the functions that people wanted from the beamformers, so we decided to provide features that would give the ultrasound companies product-differentiation. So now, it’s almost like we provide the paints and paint brushes and the palette for ultrasound companies. Then it’s up to them to decide what they want to paint on the canvas.
DESIGNING THE CORPORATE STRUCTURE
DT: So far we’ve been talking about the technology. But that’s only a part of being an entrepreneur. Tell me about staffing aa company and bringing it up to full speed with engineers and executives, and the other kinds of personnel you need to make a company function. To start with, I have this impression that Venture Capitalists don’t like founders to run a company. Does that match your experience?
AW: I don’t know if that’s true. There are four of us on the executive team and our skills complement each other. Tom Sparkman Is our CEO. Allan Evans is our VP of marketing, Richard Tobias is our VP of engineering, and I’m CTO, and the R&D guy. Allen defines the products we are going to build. Richard builds the products, and Tom makes sure that everything is running smoothly so we can pay the bills and make payroll. Richard also has done most of the hiring, since we really needed engineers. Part of our staff comprises engineers he’s worked with at at least three previous companies.
DT: Does the fact that you are CTO and not CEO make a difference?
AW: You have to know where your passion is. And also what you don’t like doing. In the early days, I did imagine myself being the CEO. But the role of the CEO is really in sales, marketing, organization, hiring, facilities, benefits, all the things that usually fall through the cracks unless you have a good CEO.
I felt that my best added value to Samplify was as a visionary, somebody who’s looking out three to five years, and asking: Where’s the technology going? Where could it be used?
Continue on next page
On that score, I also received some good guidance from Bruce Sachs at Charles River Ventures. He basically told me the same thing. He felt that my skills and strengths were mostly on the vision side. And, he’d prefer that that’s where I would stay.
DT: What would be your advice that someone who is looking to go to work for a start-up in an economy that’s just barely recovering from a deep recession?
AW: A good fact to determine about a start-up is: “How much money do they have in the bank?” That will give you an idea. . . . If you are an engineer coming into a company, it will probably take you three months to figure out is going on, and maybe another three to six months before you become productive. So you want to make sure there is at least a year’s worth of money in the bank so everyone can get their work done and figure out what the products going to do when it launches.
One of the things about being in a startup, is the freedom to try new things. In some ways you also have permission to make mistakes along the way. Samplify’s ultrasound roadmap, as I’ve described it, is an example. We started out building data converters, but later we built modules, and then we built the receive chain, and now are going to deliver the entire front end, including beam forming.
So what’s left? Some PC software on the back end, although I don’t mean to trivialize or minimize that. But we’re doing the hardware part, the real-time, embedded part, and then the part that’s left is the more sophisticated color Doppler blood flow processing.
Basically, what we’re offering the ultrasound market is lower cost, higher performance hardware that can be coupled with something as inexpensive and ubiquitous as a laptop, and that combination comprises a pretty powerful ultrasound machine.
DT: Would having put so much NRE into ultrasound mean that it was difficult to expand into other application areas?
AW: No. Our series-B funding brought in two new strategic investors. One of them operates in the seismic processing area. They have a huge bandwidth and storage bottleneck.
And there are actually three other areas outside of ultrasound where Samplify is still very much engaged with customers. For instance, computer tomography, where we’ve been working since day one, where Moog Components Group, in Blacksburg, Virginia, is Samplify’s partner. Moog makes the slip rings, the spinning donuts that revolve around the patient during a CT scan. Working with Moog is an ideal partnership, because we can integrate compression into the existing FPGAs in their slip rings.
The second market in which Samplify is quite active is in wireless base stations. Samplify filed a patent almost four years ago that said, in effect, “There’s an awful lot of sampled data going up and down the cables in cell phone towers. If you can compress at the transmitter and decompress at the receiver, AND meet the quality requirements in terms of error-vector magnitude (EVM), you can save a lot on the cost of the electric -to optical transceivesr at the receivers and transmitters, and also on the FPGAs the drive those transceivers.
I already mentioned seismic exploration. Within the seismic processing there are as many as seven different steps in which compression offers value.
Interestingly, the newest aspect of processing seismic data is supercomputing. Seismic involves the transfer of, literally, terabytes of data. You’re talking about ships recording data from 10 km-long streamers, totaling 100,000 sensors and sailing for 4 to 6 weeks at a time in order to map the sea floor. Processing that much data in a reasonable amount of time is a super-computer application that obviously has a big I/O problem. It turns out that this seismic customer can’t feed the 512 cores on the latest NVIDIA Fermi chip, fast enough. That’s where Samplify compression comes in.
So our latest thing is looking at high performance computing ,which is basically processing numbers, and in most cases, these are 32-bit floating point numbers.
As far as I can say, no one else has looked at doing lossy compression of floating-point values for high performance computing. As I look at high-performance computing, it’s just another signal processing problem, because you have sampled data flowing between processors. Why shouldn’t compression be useful there as well, as long as the quality of the results don’t change?