• As an EIR (Entrepreneur In Residence) at Benchmark Capital I started a company called ResiNet (Residential Networks – later named Travertine Systems), to develop low cost “SecondPCs” for the home along with the network to connect them. These PCs did not use hard drives, were video capable, and required a high bandwidth in-home network.
• Developed 20Mbit/s physical layer networking technology using in-home telephone wiring that was compatible with POTS (Plain Old Telephone Service) and ADSL. The system used QPSK modulation and aggressive 60-tap equalization.
• Filed 3 patents:
  1) Home Area Network (HAN) patent – the main System and Method patent
  2) Virtual Gateway System and Method – a patent leveraging the Frequency Domain Multiplexing (FDM) design of the network to allow POTS, ADSL, and the HAN to coexist on the same twisted pair wire while maintaining POTS lifeline service
  3) A Filter patent for a simple passive (Butterworth type) filter than could be user installed on home telephone wire to improve spectral isolation between ADSL and the HAN while maintaining electrical connectivity essential to POTS lifeline service
• Pitched Travertine to a number of local Tier A VCs to raise a Series A round
• Sold the Travertine assets to Epigram, another Benchmark company, in late Oct ’97. Epigram was to refine and build upon the Travertine HAN technology and then contribute it to the HPNA (Home Phone Networking Alliance) as the basis of the HPNA 2.0 standard. Epigram was acquired by Broadcom in April ’99.

EIR at Benchmark: Just before departing Chromatic Research, I was invited by Mike Farmwald (founder of Rambus and Chromatic Research) and Bruce Dunlevie (founder and general partner at Benchmark, and a lead investor in Chromatic) to spend time at Benchmark as an EIR (Entrepreneur in Residence) incubating an idea that Mike had for the creation of lightweight “SecondPC” product(s) for the home. By lightweight, it was meant the product(s) would not have a hard disk drive, and so was an early example of an internet endpoint that was dependent on the network to operate – including potentially paging Windows over the network. At that time the majority of PC purchases were still the “first PC” in the household, and Mike was anticipating lighter weight and less expensive products to follow the penetration of the “first PC”.

I took up residence at Benchmark in mid April ’97. My engagement was a fairly classic EIR model, where you either come in with an idea to be developed, or are expected to develop an idea or bring in a startup. You were paid a stipend during a typical stay of 6-12 months. The sponsoring firm had right of first refusal – and if they did not want to fund your deal you were free to take it elsewhere. There were no formal documents to this effect – this was just the general understanding.

The In-Home Network was Key: A key goal of SecondPC products was supporting video over a network in the home – which translated to a data rate of at least 15Mbits/s as MPEG-II main-level main profile was 15Mbits/s, and Advanced Television (ATV) was slated to be 19.2Mbits/s, so I quickly settled on 20Mbits/s as the network design target. The first issue was what network to use, and since WiFi was not yet ratified, and WiFi network products running at only 2Mbit didn’t start shipping until 2000, an alternate network had to be found. It was the network that quickly became the primary product focus of Travertine.

In 1997 the primary source of video from outside the home was either cable or satellite. The first ADSL (Asymmetric Digital Subscriber Line) standards using phone wire to the home were not approved by standards committees until ’98 and ’99 and took some time thereafter to be deployed – and these first generation deployments in practice supported data rates well below what were needed for most users to support anything other than browsing or audio streams.

The primary source of video inside the home was still VHS tape, although DVD based video was rapidly gaining traction.

Use In-home Phone Line: Since stringing Ethernet cable around the home was not an attractive product feature, the thought was “could we use in-home telephone wiring (UTP – Unshielded Twisted Pair) to create a network”? I set about investigating such a network (which I called the HAN, for Home Area Network) using the company name ResiNet (for Residential Network).

Travertine still pursued product definition of SecondPC products, lightweight PC like appliances that would lean heavily on the network. It was desirable that these types of products booted very quickly (“instant on”) and Jay Pattin, who had previously worked with Mike, was brought in to look at SecondPC type product software issues.

Working at Benchmark: A Tier A VC firm like Benchmark is a marvelously productive place to explore new business ideas and to develop technology. For one thing, if you called someone – they would actually call you back! – and usually right away! Even someone like John Cioffi! Furthermore the close proximity to Stanford facilitated fast frequent trips to the engineering library to consult texts and journal articles (I know its hard to imagine – but not everything was on the internet back then).

Epigram and V.IP: Mike was also involved at that time with a Benchmark funded startup called Epigram (Mike was considered a founder). I was told at the start of my engagement at Benchmark that Epigram was primarily focused on the last mile and a Layer II/III protocol they called V.IP – which used Internet Protocol and was packet based down to Layer 2 to deliver “Internet Dialtone”. Mike (and Benchmark) of course wanted ResiNet to use Epigram’s technology for delivery of data to the home over “last mile” phone wire, and wanted SecondPC products to support V.IP.

ResiNet Focus on Physical Layer: The TCP/IP (internet) reference model uses a 4-layer stack, with Layer 2 as the Internet Layer (equivalent to Layer 3 of the OSI model), and Layer 3 as the Transport layer (roughly equivalent to Layer 4 of the OSI model).

The networking technology that ResiNet developed for home wiring dealt solely with the lowest layer – the Physical layer. Once the physical layer is in place – the higher layers can be installed above it with but a few caveats – with one of those being whether the physical layer readily supports collision detection as used in the standard CSMA/CD (Carrier Sense Multiple Access with Collision detection) ethernet protocol. Resinet was leaning against CSMA/CD for reasons which are presented in more detail below in the downloadable Traverting Home Area Network Technical specification, and was instead considering token based TDMA (Time Division Multiple Access) protocols for the MAC (Medium Access Control) layer as they could be more efficient (especially under heavy loading) and provide better Quality of Service (QOS) guarantees which was deemed quite important for video delivery over the Home Area Network.

I never learned any technical details of what Epigram was doing – Mike was pretty conscientious about maintaining a “Chinese Wall”, and I learned more from emails sent to me by Epigram folks than anything else – such as mentioning “constellation sizes” etc when describing their modulation approach – which was reasonably consistent with a focus on the last mile.

In-Home vs Last Mile Wiring – Very Different!: The physical, electrical and signal transmission characteristics of the twisted pair wiring inside the home is very different from the POTS “last mile” wiring between the home and the exchange or the DSLAM (Digital Subscriber Line Access Multiplexer). The wiring inside the home is very unconstrained by comparison, with the in-home network containing arbitrary numbers and lengths of branches, necessitating quite different physical layer modulation and error correction approaches. Wiring outside the home over the “last mile” was heavily studied and characterized (as it is the environment for both POTS (Plain Old Telephone Systems), standard modem (such as V.34), and xDSL delivery) and was limited to (at that time) frequencies below 1.1Mhz primarily because of very strong attenuation (loss of signal) for higher frequencies given average subscriber line or “loop” lengths of over two miles in the US. ADSL itself has a maximum “range” of a little over three miles before a repeater is required. Since ResiNet’s goal was to support multiple low cost consumer products on the home network, the power and compute resources were more limited than what a consumer might be willing to pay for a single DSL modem or set-top box.

Unshielded Twisted Pair (UTP) telephone wire in the home is typically “unbalanced”– meaning the nice regular twisting normally found between the wire pairs outside the home was disturbed because of wiring to RJ-11 jacks, wiring to stubs, yanking through wall studs by contractors, etc. This unbalancing causes excess energy to be radiated, creating interference between neighboring wire pairs and possibly exceeding the amount of allowed radiated RF energy under FCC part 15 (potentially interfering with the AM radio in the kitchen, etc) – which in turn limits the amount of energy you can transmit with, which in turn limits the maximum data rate of the network. Furthermore, there could be arbitrary wiring topologies – such as rings, or stars, or trees – all of which created loss of signal energy (called nulls) at frequencies that were dependent on the physical topology (layout). Also any time someone took a phone off the hook, or if someone plugged a phone into an RJ-11 jack somewhere, the electrical and spectral characteristics of the in-home UTP network would change. Noise from HAM radio sources or impulse noise from electrical motors in the home also had to be anticipated.

Huddling with the Experts: I immediately set about familiarizing myself with the relevant standards requirements starting with Part 68 Title 47 of the CFR (Code of Federal Regulations) as well as various TIA (Telecom Industry Association) standards. I aggressively engaged with with communications and modem/DSL consultants and industry experts such as James Long, Mark Flowers, Whitham Reeve, Brooks Read, Steve Bradley, Kim Maxwell, Walter Chen, and others – and these engagements proved crucial to rapidly converging on an optimal approach. Fortunately – as I had recently completed the design of a V.34 modem interface and DAA (Data Access Arrangement) as part of a PCI board I had designed at Chromatic Research – I was already up to speed on most of the standards nomenclature.

Got to Have Some Empirical Data: To familiarize myself with electrical, spectral, and topological characteristics of typical home phone wiring I rented some Tektronix test equipment and characterized some homes, starting with my own (a drawing from that exploration is in the slide show below), and then I moved on to the office at Benchmark and another half dozen or so homes of friends. Not exactly an exhaustive survey, but enough to be quite informative.

Early Discussions with Epigram: In a meeting with Ed Frank at Mike Farmwald’s house at the end of April we talked about cooperation between the two companies. Ed said that “Epigram’s goal is to make a number of different physical modulation schemes flourish under V.IP” and that “V.IP was primarily a Level II protocol and the crown jewel and the company destiny came from control of the Level II protocol” and that “If what we (ResiNet) do is to figure out how to make V.IP work at 20Mbits in the home – then great!”. If we were to license V.IP – his licensing stance at that time was on a per client (anything with a V.IP modem) basis. Afterwards Mike stated his preference was a royalty free licensing model from Epigram – implemented as an equity stake in ResiNet along with a straight cross-license agreement with no right to sublicense or transfer. In a phone call with Ed a few days later (4/29) he said he liked the idea of an equity deal as it would establish a precedent for the “value” of V.IP.

In a breakfast a week later with an old friend from Chromatic Research (Ali Alasti) he threw some water on my optimistic expectations for progress on our licensing discussions with Epigram by saying “why would they want to grant you a license to V.IP when home networking is probably on their roadmap”.

ResiNet becomes Travretine: Our lawyers at Wilson Sonsini called me on May 6 to tell me there were too many conflicts to use the name ResiNet. I thought of alternatives and a few days later settled on the name Travertine Systems. I just thought Travertine was a nice name, and that the patterns in the Travertine marble were reminiscent of the kinds of arbitrary wiring topologies one saw in home phone wiring.

Why not Use FDM (Frequency Division Multiplexing)? In a meeting with Brooks Read (who was a modem jock then working at Chromatic Research) on May 7 we discussed the possibility of putting the HAN (from a frequency perspective) above DSL – effectively employing coarse Frequency Division Multiplexing (FDM). Since the overall wiring length in the home was much shorter than the loop length of the “last mile” – that attenuation at the higher frequency would not be nearly as severe – and there were many other advantages including:

• Allowing compatibility with any (other) xDSL scheme
• Allowing better QOS on the HAN as there would be no interruptions by transmission on the xDSL or POTS channels
• Insuring compatibility with lifeline POTS service (there must be a DC current path)
• Eliminating the need a POTS splitter (which was huge)
• Decoupling us from the Epigram V.IP phy (which was almost certainly very different as it was then focused on the last mile and needed to maintain compatibility with xDSL frequency ranges).
• Eliminating the need to synchronize the domains to transfer packets
• Eliminating the need to “snoop” the subscriber line to transfer data in the premises
• We could isolate the frequency domains using a simple inexpensive passive RF filter

A diagram of the FDM spectral allocation for the ResiNet HAN is shown below. The diagram shows the HAN spectrum starting at the top of the DSL band at 1.1Mhz, but in practice the HAN started in the 3-4Mhz range to enable the use of low cost passive filters to provide better isolation between the two bands.

Neal Margulis Joins the Team:  In early May Neal Margulis joined the Travertine team on a part time basis.  Neal had been at S3 in a Sr. management role and after Travertine became CEO of iCompression (where Benchmark was an investor).  Neal was a more seasoned senior exec than I, so Neal took the CEO title and I would be founder and CTO.  Neal was actively involved over the next five months with emphasis on business models, licensing issues, SecondPC type product ideas and priorities, attending funding pitches with me, etc.  I focused on the HAN technology & implementation, and and all the legal/patent tasks.

“Business Model Expansion” at Epigram?:  In mid-May Mike and I met with Ed Frank et. al. at Epigram   They presented a suite of products called EPIGATE, EPISWITCH, UNIC (Universal NIC), and EPINET (this was the first time I had heard about any of this).  And now, two weeks after the meeting with Ed where he said “V.IP is the crown jewel that defines the company, this had changed to “we are a networking company” and “EPINET defines the company”.    He also said that Andy Rachleff (the Benchmark partner that invested in Epigram) had told them that “the company is not fundable without a networking component”.

A few days later during a meeting with Bruce Dunlevie he said that “Epigram did not yet have a sense of what business they are in” and they are in an “expansionist phase of their business planning”.  Bruce had hoped that over the last week or so they would have converged on a plan – but that didn’t happen – perhaps because of the potential arrival of Frank Marshall (ex Cisco) as their new CEO.

Travertine Patent Work Begins:  At the end of June Travertine was incorporated and in early July I started meeting with lawyers at Fenwick and West to begin patent work on three patents:

• A main Home Area Network (HAN) systems and method patent,
• A patent on a Virtual Gateway concept (which leveraged the use of separate frequency ranges – FDM –  for the different physical layers). The patent addresses how POTS, ADSL, and the HAN can all co-exist on the same twisted pair, and communicate between each other, while continuing to support POTS lifeline service – meaning the twisted pair has to stay electrically contiguous to pass DC current.
• A patent on a passive clamshell filter which improves isolation between the HAN and ADSL frequency ranges on the same twisted pair that does not break the DC current path to the central office.  The rolloff or “skirts” of this low cost passive filter is why the HAN frequency ranged needed to be 2-3Mhz above ADSL.  This type of patent can seem incidental – but this sort of nuts-and-bolts technology can be crucial to commercial viability.

Modulation Type Selection:  One of the most important design decisions I had to make was the type of modulation approach used to transmit data on the HAN.  Key requirements were simplicity, low cost, and robustness – while still meeting a 20Mbit data rate in the available (~16Mhz) spectrum in the face of the significant dynamic impairments of the home networking environment while accounting for overhead for error correction and other factors.  I settled on QPSK (Quadrature Phase Shift Keying) for reasons that include:

1. QPSK only requires a 6-bit Analog-to-Digital (A/D) converter.  Which enables use of less expensive A/Ds that can still sample at rates fast enough to handle 20Mhz modulation (~60Mhz sampling rate).
2. A much lower Signal-to-Noise (SNR) is required to recover data vs more complex modulation schemes (~10db less than QAM-16).
3. Automatic gain control can converge much more quickly, minimizing the amount of wasted bandwidth spent on preambles.
4. Signal recovery is much simpler (just a hard limiter is needed) for this simple modulation scheme, vs higher order modulation – which adds another 6db of noise immunity from deterministic noise sources.

Aggressive adaptive equalization was required to compensate for channel impairments such as multipath echo.  Since between every transmitter and every receiver on the network the channel “looked” quite different from an impairment (spectral response) standpoint, different equalization was required for each transmitter-receiver pair.   To provide 1.5us worth of equalization at a 4x sample rate a 60 tap filter was required.  This was doable in 0.25 micron silicon.

An example of what a Travertine Home Area Network might look like is shown below:

Travertine’s First Products:  Travertine’s first products then would be PCI and USB Network Interface Cards (NICs) that would be capable of putting out video.  The first generation would use off-the-shelf third party baseband silicon from providers such as Stanford Telecom (whom Travertine had extensive discussions with), while next generation products would utilize baseband modem chips developed by Travertine – enabling  Travertine to potentially choose a board business model, a fabless semiconductor business model, or both.  Example diagrams of both first and second generation PCI based boards are shown below.  We also wanted our technology to be integrated into DSL modems in order to bridge from the DSL physical and MAC layers to what would be used by Travertine’s HAN.

In a meeting on 7/8 with Mike he told us that “Epigram will license freely – but you must be compatible with their physical layer” – which of course made no sense for Travertine unless Travertines primary business model centered around in-home networked appliances (“Second PCs”) as it appeared Epigram’s biz model was becoming more and  more in-home centric.  Mike said, “you were free to improve as long as you maintained compatibility with their (Epigram’s) core”.

Benchmark Advocates Focus on the Network:  A few days later in a meeting with Bruce Dunlevie and Neal Margulis, Bruce thought we should focus on being a networking company, and defocus any work on “SecondPC or netPC”.  One reason he gave is that there is insufficient return on invested capital for low margin products like SecondPC – “just look at Dell” he said.  Bruce wanted a biz plan (headcount, budget, etc) from us for presentation to the partnership at the end of the month

On meeting on 7/17 with Bruce, Neal, and Mike – Mike continued to push SecondPC type products – that Travertine should design, build, and ship SecondPC boxes to OEMs. (Mike would agree several months later later that the SecondPC idea had little value).

Epigram Changes primary focus to “Residential Ethernet”:  A few days later there was another meeting with Ed Frank at Epigram.  Now the corporate objective is “Residential Ethernet” (over telephone wire).  It is to be an open standard – and “they will embrace whatever modulation technique is required to get to 15-20Mbits/s”.

On 7/20 Neal, Mike and I meet with Bruce Dunlevie.  The consensus on the value proposition was: “We bring networking technology to the market” where SecondPC is a good “first deployment”.   There were concerns expressed that Epigram’s new market focus would limit Travertine’s opportunities and preclude long term value.  Bruce observed that if “3COM, MSFT, and Intel all buy into V.IP and push it there will certainly be a roadmap to higher bit rates inside the home while will put strong price pressure on Travertine’s IP”

On Tues 7/24 Neal and I presented the Travertine opportunity to the full Benchmark partnership.

Kevin Harvey (Benchmark) advocates defocus on the SecondPC Type Products:  In early August Neal, Mike and I met with  Bruce and Kevin Harvey (Kevin was a Benchmark partner).  Kevin had never seen the pitch.   After the presentation he said we should discard SecondPC – that from a customer relationship as well as marketing standpoint it was problematic.

In mid August Jeff Thermond arrived from 3COM as Epigram’s new CEO.

Funding Pitches to other VCs:  During the first weeks of August, Neal and I pitched to Mayfield Fund (Bill Unger and Kevin Fong),  Sequoia Capital, and USVP (Irwin Federman, Mark Friend, and Stu Phillips) and one or two others.  Mayfield wanted to seed the company to validate the technology and arranged a follow up due diligence meeting at Cisco that was attended by Bill, Kevin, Neal, me, and Luca Cafiero and Mario Mazzola of Cisco.  Luca and Mario came to Cisco through Cisco’s acquisition of Crescendo Communications in ’93 (Crescendo had developed FDDI-speed LAN over unshielded copper twisted pair wire – and was funded by Mayfield), and when they both recently left Cisco last  year Luca was SVP and GM and Mario SVP and CDO.

Proverbs from Irwin:  Irwin Federman was an impressive guy then and undoubtedly still is.   His biggest concern was that Travertine did not yet have a strong OEM endorsement.   And he was quite effusive with some choice advice, including:

• “We are like butchers here – we see a lot of beef go by – and we know the difference between choice and prime”
• “We don’t raise the beef – about all we can do is occasionally cut straight”
• “Are you guys sure you want to do this? For us its only money – for you it’s the filet of your life”
• “We need to be sure there is enough of a there there”
• “Sell us – convince us of the error of our ways”
• “Don’t aim low”
• “You ought to think about enabling something entirely new – not just making something a little cheaper”

Needless to say I think Irwin thought this deal leaned more toward choice.

Benchmark and Farmwald Bow Out:  By mid August Benchmark had declined to participate in Travertine citing conflict of interest with Epigram.   Mike had stepped away from Travertine as well via assignments of IP and stock to the company.

Epigram Decides to Push V.IP as an Open Standard:  In a meeting with Ed Frank on 8/18, Ed said that Epigram now focused on making V.IP a standard.

Acquisition Discussions with Epigram:  Discussions began in mid August about the possibility of Epigram acquiring Travertine, and negotiations accelerated during the next month such that an asset sale of Travertine to Epigram closed on Oct 21, 1997.  Approximately 18 months later, after morphing into a fabless semiconductor company developing chips primarily around the IP Epigram contributed to create the HPNA (Home Phoneline Network Alliance) 2.0 standard,  Epigram was acquired by Broadcom.

TAKEAWAYS:

Ironically, Epigram evolved in to a fabless semiconductor company – they (so far as I know) did not build and sell any central office hardware or products like the EPISWITCH, or the EPIGATE, or the UNIC.  They were not to implement “Internet Dialtone” in the sense of providing any last mile technology or implementations.   What they brought to the party w.r.t. Broadcom was chip level implementation of their residential network, including the physical layer, which they had contributed to the HPNA (Home Phoneline Networking Alliance) as the basis for the HPNA 2.0 standard.  The physical layer of this contribution looks a lot like an evolved version of the physical layer acquired from Travertine.

Benchmark had already made an investment in Epigram and so when Epigram’s business model evolved (“expanded” to use Bruce Dunlevie’s term) such that it came into greater conflict with Travertine’s in-home network physical layer focus, their “hands were tied”.  Nothing personal – just business.   Mike and Bruce had tried to make my experience at Benchmark a pleasant and productive one.  I greatly admired the Benchmark team.

Perhaps only 1 in 50 startups “make it big” – so every promising opportunity is precious and you should fight hard to hold on and make the most of it.  But then again sometimes you just have to roll with the punches – the opportunity at Benchmark was too good to pass up.

RESOURCES:

EPILOGUE:

Shortly after departing Benchmark Bruce and Mike asked me if I would like to join Matrix Semiconductor.   Matrix was developing a novel 3D memory technology that initially was Read Only, but there was hope at that time that it would be capable of both read and write (which did not pan out).   Matrix’s investors included Benchmark and Skymoon (Mike’s boutique VC firm).  While I much liked the idea of working at a company where Bruce and Mike were both involved, I was wary of plans that involve developing a new semiconductor device and/or process – as it always takes more time and money that one originally hopes.  I was also by then becoming involved in a Software Defined Radio startup (that became nBand Communications) that would blend both my interests in communications and my interests in processor  architecture, see NBAND-COMMUNICATIONS.  Sandisk was to buy Matrix in 2005 for ~$238M, and later was to abandon the technology as improvement in multi-level Flash was simply too relentless.

The challenges facing startups  developing new semiconductor processes/devices or attempted to “tweak” a main foundry process such as TSMCs would regularly show up in subsequent years, including with startups funded by Tallwood when I was there, as well as for well funded companies such as SuVolta.  Tallwood courageously attempted to address the dwindling options for companies looking to develop novel devices, processes, MEMS devices etc by creating the SVTC (Silicon Valley Technology Center).   But unfortunately SVTC did not survive – leaving only those resources available at large leading academic institutions – and as they are primarily research and grad student teaching facilities, the quality control and repeatability is generally not adequate for companies trying to develop commercially viable product.

About 6 months later I met Bruce for lunch at the Stanford Golf course cafe (where he apparently liked to eat) and he shared some advice in terms of founding teams and institutional investor fundraising.  He also told me about their investment in eBay and said something to the effect that he found it amazing the IPO underwriter types it was currently valued it at over $1B.  I am grateful to Bruce for taking the time to coach me on those challenges.

Travertine Slide Show:

1. Foley Home Phone Wiring Spectral Response
2. Foley Home Phone Wiring TDR Response
3. Main Home Area Network (HAN) Patent
4. Virtual Gateway Patent
5. Passive Filter Patent