High-tech

Due Diligence: picoChip

By Bipin Parmar

1. Introduction
2. picoChip - the brass tacks
3. Value proposition
4. Dynamics of the basestation market
5. The Chilli perspective

1. Introduction

picoChip is a fabless semiconductor company based in Bath, UK, targeting the 3G basestation market with a system solution including a chip, design tools, system software and reference designs. The Chilli profiles picoChip with a view to assessing its potential for success, and makes a number of recommendations.

2. picoChip - the brass tacks

Founded in September 2000 by Peter Claydon (ex Marconi, Connexant and Oak) and Doug Pulley (ex Vodafone, Lucent, Oak), picoChip was bootstrapped by its founders and early employees, who lived off their savings. The company received first round funding of $7 million from Pond Venture Partners and Atlas Venture in June 2001, and from that point, grew its headcount from 15-20 staff to 42 today, including several non-engineering staff (admin, part-time cfo, etc), and the rest split equally between systems, tools and silicon specialists. Interestingly, two thirds of the employees have worked together before, including four of the seven-strong executive team, who were at Oak and Conexant. The company acquired systems expertise from the recruitment of several 3G system architects from Lucent and Fujitsu.

Rodger Sykes, formerly of TI, Philips and LogicVision, joined as ceo in July 2002. Rupert Baines (ex Analog Devices, Atlantic Telecom, Pond Ventures), joined as vp of marketing in September 2002.

As is typical in a start-up, employees work all sort of hours. Interestingly, most employees are supplied with laptops and broadband connections and are able to work from home. All are available on mobile phones, and calls are patched through via the company switchboard.

Baines stated that the company is pre-revenue, and declined to give a figure for burn rate, but he did confirm that picoChip has achieved sufficient traction for its next round. In terms of revenue outlook, Baines stated "We expect to see some tools revenue in 2003, with volume business in 2005."

3. Value proposition

Before discussing picoChip's technology, it is worth reviewing the processing tasks required for a wireless basestation. A typical basestation design requires radio frequency (RF) and power amplifier expertise, as well as a high-speed baseband, executing specific DSP operations and complex control protocols. Baines states that, "The MIPS effort for 3G is 500 times harder than for GSM." Basestation solutions have traditionally used custom ASIC/FPGA technology together with DSP and embedded processor devices. The designer has to trade off between cost of the components, form factor, power consumption, development complexity and overall development cost. picoChip has developed a baseband device that is fully software programmable, allowing basestations built around it to be remotely reconfigured and upgraded as 2.5G/3G network capabilities evolve.

picoChip's device, the picoArray, is a massively parallel array of heterogeneous processors. On one die are 430 discrete 16-bit devices with their own data and program memories. Each element's horsepower is similar to an ARM9. A deterministic fabric interconnects the processor elements. These individual elements come in several different flavours, with elements supporting two distinct classes of operations, e.g. dataflow and control. picoChip claim performance of 30 GMAC/s (giga multiply and accumulate operations per second) at a clock frequency of 160MHz. To put it in some kind of perspective, this is several times the performance of the Texas Instruments (TI) TCI100 DSP, which clocks at 720MHz. This should give picoChip a strong advantage in power consumption over both TI and the Analog Devices (AD) TigerSHARC, although system vendors would look at the complete system, including rf and power amp. Power dissipation is important, as thermal issues limit the number of channels that can be accommodated on a single linecard. A solution with lower power would allow an increase in packaging density. According to Baines "picoChip aims to provide up to 90% power saving per channel over existing solutions."

picoChip claims to have developed a redundancy technique that enables software to be remapped onto different processors. This enables picoArray devices with some failed processors to be used, increasing effective yield. This should allow for competitive pricing against similar sized devices. Working samples, of a picoArray in 0.13mm from TSMC, are currently being evaluated by a tier one telecom equipment vendor.

The development tools suite for picoArray consists of an ANSI C compiler, debugger, assembler, linker, simulator and a place and switch tool that parallelises the code prior to compilation. The tools were developed as part of the overall system approach, rather than as an afterthought, as is the case with most platform solutions. With a single picoArray device potentially replacing ASIC/FPGA, DSP and embedded processor devices, the development complexity is reduced with several tool suites replaced by one. The resulting form factor of the end product is also reduced.

picoChip supplies valuable system software with a rich blend of basestation library functions as product-level code. The guarantee of functionality depends on each business case, and depends on the level of test and verification required. According to Baines "New entrants will take all steps required to gain rapid market entry, and would pay for the code and consulting support, whilst tier one vendors have their own in-house value to add, e.g. rake receivers, etc."

The picoArray device is positioned as software system-on-chip (SSoC), with developers able to add functionality and differentiation by placing software onto groups of processors. Baines asserts, "We believe that our unique system architecture, allowing customers to easily add value through software differentiation, will allow us to command better than average margins."

4. Dynamics of the basestation market

So how does picoChip's proposition stack-up against it's target market? Lets first take a look at what is driving the market.

There is a definite need for 3G, with existing 2G/2.5G networks becoming capacity-bound. The important question is, 'when will 3G infrastructure really get deployed in volume?' In a way, the delay has worked in picoChip's favour, as it opens up another window for next generation, lower cost designs. Mobile network operators (MNOs) are seeing a continuing decline in average revenue per user (ARPU), due to competition, saturation and price reductions on the gravy train areas such as call termination charges to fixed networks. The operators will have to find alternate sources of revenue to balance this decline, if not grow. Data in all its forms, whether text, pictures, video and music, is the next growth area for operator revenue. According to picoChip, 40,000 3G basestations were installed last year.

The cost of basestation development, according to Baines, "is in the region of $100 million". According to Baines, "Equipment vendors promised low costs to the operators, and are now contractually obliged to supply at that cost. It has been estimated that the cost of sales for one leading equipment maker is 126%, so they, and other vendors, are losing significant amounts of money per basestation." The baseband processing capability (ASIC/FPGA, DSP, embedded processor) is currently 20% of the silicon content of a basestation. PicoChip claim to be able to replace twenty DSPs with one of their devices, and claim that a 64 channel base station could be built using only eight picoChip devices, with an 80% cost saving on the bill of materials (BOM). A picocell could function with perhaps one chip, according to Baines, "The current BOM for a basestation is approximately $100 per channel and picoChip are currently in the $10-30 per channel range, with a roadmap to going under $5 per channel, when volume starts"

According to figures from Morgan Stanley, 2002 saw a $5 billion investment in 3G infrastructure. Figures from In-Stat, indicate that basestation semiconductor revenue is expected to be $2.48 billion in 2003, with the forecast for 2006 dropping to $1.41 billion, due in part to factors including downward cost pressure, the spectral efficiency of 3G over 2G, and the increased integration of functionality.

5. The Chilli perspective

Applying The Chilli Value Test raises a number of business challenges to be addressed by picoChip's management.

Technology vendors must present their solutions on the basis of tangible value added to the end customers business, and think more in terms of metrics such as total cost of ownership (TCO), total bill of materials (BOM) and reduction of business (not development) risk [Editor's note: more on these topics will be appearing in future issues of The Chilli].

In terms of the product proposition, picoChip's SSoC concept is very close to The Chilli's definition of a platform. For this to be a successful strategy for picoChip, the following criteria would have to be developed further:

• Opportunity to get cost-down via integration of DSP and FPGA logic and removal of NRE cost (ASIC), so there should be multiple ASIC, FPGA and DSP sockets to replace. This has to be clearly demonstrated in terms of figures for chief financial officers at prospective customers
• Differentiation can be limited to software enhancements
• Domain expertise in the end application
• A complete system built around picoArray

picoChip must hide the complexity of parallel processing 'under the hood' and focus on sub-system solutions. They have taken the correct steps in that direction by demonstrating a sub-system design including a radio interface. An off-the-shelf standard product is easier to design with than semiconductor IP or a custom ASIC, with no issues regarding NRE costs or verification costs. If there is a truism for the basestation silicon market, it's that it's a systems business. New entrants cannot survive by piecemeal solutions alone. This means that beyond a baseband chip, a serious solutions provider must consider the antennae, radio, power amplifier and power supply. The picoArray product must therefore be offered as a part of a sub-system solution rather than a semiconductor device in its own right.

The sub-system solutions cannot be resourced completely by picoChip alone, so a clear partner strategy, with appropriate management time, resource allocation and attention will be required. According to Baines, "We are conducting an early exploration of partnerships with radio vendors." picoChip has remained very focused in its choice of market, but requires partners who are strong in the non-baseband areas, ideally those who are already suppliers to the existing basestation equipment vendors.

picoChip should already be aware of customer culture. The 'not invented here' syndrome cannot be underestimated. While the equipment makers are downsizing, they will of course focus on core competencies. For some vendors, this may include the design of the basestation product platform. Beyond the development team, there is also the field service organisation, which cannot be laid off despite remote reconfiguration, as visual inspections and environmental maintenance will still be required. Furthermore, conservatism should never be underestimated. Deployment of network infrastructure encompasses planning permission, penalties for delay, maintenance schedules, etc.

There are successful, long-term relationships between incumbent suppliers and equipment makers. Both sides would be loath to forego these, unless there is a significant pain point being addressed by a new entrant. In this kind of market, a start-up will need a good partner network.

picoChip faces stiff competition from incumbent suppliers such as Texas Instruments (TI) and Analog Devices (AD) and other start-ups attacking 3G basestation sockets. TI currently claims to have eight out of the top ten operator sockets. It is not clear if these are evaluations or solid design wins. Another UK-based start-up, Aspex Technology, with its massively parallel technology, is also eyeing this market, as are Elixent, PACT XPP and Morphics, recently acquired by Infineon. High-end, massively parallel technology has stumbled along, from 3D graphics/virtual reality to network processing/cryptography, through DSL and now onto basestations, with each segment held by a number of commercially strong incumbents. Future segments for many of these companies would include wireless metropolitan area networks (IEEE 802.16) and mobile broadband wireless access (IEEE 802.20). So picoChip is in a fortunate position as it has the ability to diversify into other markets.

Comments on this story? Send an e-mail to Woz@TheChilli.com or Bipin@TheChilli.com

© Chilli Publishing Ltd 2003

17MAR2003