Tuesday |
|
|
| Interview |
Interview with Dr. Paul Meissner, CEO and President, Santur CorporationIntroductionSantur, founded in 2000, is a major supplier of PICs (Photonic Integrated Circuits) for tunable transport in optical networks and for high bandwidth extended reach data links. The company's Intelligent IntegrationTM technology is based on core expertise in laser arrays, semiconductor modulation, device packaging and high yield volume production technology. Santur claims to supply the majority of the world's 10G, 40G and 100G tunable sources, and is applying Intelligent IntegrationTM to its next generation of lower cost, lower power consumption, smaller footprint devices. In addition the company is using its arrayed laser and packaging technology for non-tunable applications, specifically the development of 40 and 100+ GHz transceivers and components for high bandwidth aggregation and interconnect applications. The following is a discussion around a presentation given by Paul Meissner at the OIDA forum entitled 'Fabrication challenges and opportunities in photonics', held March 10-11, 2010. [Download slides accompanying the OIDA presentation (pdf).] Background - from Applied Materials to SanturPrior to Santur, Paul Meissner worked for a decade in the semiconductor equipment business with Applied Materials and KLA-Tencor. Much of his experience in device manufacturing and foundry operations came from his experience at Applied Materials, focused on high-end single wafer thermal processing: "The aim of this work was to replace large batch furnaces supplied by companies such as Tokyo Electron and Kokusai - a difficult challenge as batch furnaces were a proven and robust solution". As part of this work Dr. Meissner's team investigated the significant differences between major high volume, low mix, integrated device manufacturers - such as Intel and Samsung - and relatively low volume, high mix foundries used by fabless chip companies that typically required runs in the order of only 1,000 wafers, taking less than a day's capacity. In the high mix foundry environment, the need to batch wafers before loading them into furnaces was seen to cause dramatic drops in cycle time and expensive increases in WIP (Work in Process) inventory that could impact profit, and more importantly, affect the guaranteed time of delivery to the customer. One incentive to manage the operation effectively, is that a foundry can charge a premium for quick delivery to the customer. Dr. Meissner noted that as a result of this work, Applied Materials developed expertise in modelling fab layout and developing a compelling value proposition for foundries based on moving to single wafer production. By running the line 'lean', expenditure on equipment could be significantly lowered. Subsequently, Applied Materials created a development services model showing customers that the furnace could be replaced. To do so, the company targeted the most challenging application, the gate dielectric, key to the operation of the transistor, and the most important device feature to achieve the scaling required by Moore's law. As gate scaling went from 40 down to 20 Angstroms, and on down to 12 Angstroms, it became more and more difficult for companies to stay on the Moore's law time-scale since the development of a production process required perhaps a year of testing out equipment before a new transistor could actually be designed and produced. Describing Applied Materials's solution, Paul Meissner said the company sought to offer a working system but to do so required working transistors to demonstrate it, "To achieve this we either had to build a $100 million fab or outsource the work to another company. For this application, the team went to Hitachi, which agreed to deliver processed lots with all the electrical parametric data that it ran for $25,000 per lot". Via this approach, Hitachi delivered pre-gate wafers to Applied Materials, which then processed them under different conditions before sending them back to Hitachi for finishing and testing. Applied Materials was then able to go to customers with the functional transistors together with all the parametric data. Santur has its own Indium Phosphide device fab in Fremont, California, however according to Dr. Meissner, Santur sometimes utilises this same model for some of its products. In certain cases, Santur uses outside facilities to develop new device types and ensure that they work before bringing them into its own facilities for full production. As a result, the decision on how best to invest becomes a tactical rather than a strategic one. As Dr. Meissner puts it, Santur basically operates like an integrated device manufacturer (IDM) from the silicon world, "Santur has its own InP and silicon fabs colocated with our packaging pilot line. In the silicon world, some IDMs outsource packaging, others don't, but whatever the case, the packaging production is all done in Asia. As an example, Intel has its own packaging facility in the U.S. where the process is developed before being put out for volume production in Asia". Optical components sectorIn his presentation at the OIDA event, Paul Meissner began by noting that tech stocks peaked 10 years ago and asked, where is the industry now? Some would reply, he said, that the optical components sector is still working through the wreckage of the crash of 10 years ago, others that it is currently being eaten alive by lack of profitability. When speaking to investors on this topic, Dr. Meissner noted, the comment is often made that stocks in this sector are not bought to hold, they are bought to trade, as the industry is plagued by what is termed 'profitless prosperity'. This is a key concern for the sector because it needs to attract investors and investment, particularly in small, innovative companies, to be able to build a profitable business based on the clear demand for bandwidth: "After all, the industry is enabling provision of high bandwidth coverage in developed nations and cellular service in developing countries, both of which rely on low-cost telecom equipment - this is the 'prosperity' part of the phrase cited above". According to Dr. Meissner, the industry therefore needs to address the profitability issue by looking at the fundamental structure of the business, "I would say that deciding, for example, whether to be fabed or fabless is a relatively minor factor that is unlikely to have a substantial effect on whether a company in the communications optical components sector is profitable or not". To make his point, Paul Meissner set out to evaluate the impact of different factors on the silicon IC industry, including the adoption of the fabless model, to see what could be learned about how to best achieve profitability. Fab or no fabAs part of the study to assess the relative merits of owning or not owning a fab, Dr. Meissner used a PriceWaterhouse report that compared a number of IDM and fabless companies over the period 2005 to 2009, with the companies divided into three groups - large, medium and small. For the purposes of the study he focused on the small/medium groups as there are no large - $750 million-plus annual sales - players in the communications optical components sector. Additionally, such groups include a lot of fabless firms. The first point shown by the study was that in general there is a clear distinction between small/medium IDM and fabless companies, with small fabless firms generating average GPMs 9 points higher than IDMs and mid-size fabless firms achieving GPMs 12 points higher. However, Dr. Meissner noted that there is a very wide GPM range for the companies covered in the PriceWaterhouse report, with GPMs for fabless companies going from 0 or negative up to around 70% for Mellanox. In other areas IDMs generally do better than fabless companies, he found, notably with regards to R&D/engineering and SG&A costs. In these areas, although small IDMs do not fair much better than fabless firms, medium-sized IDMs tend to do significantly better in terms of R&D and SG&A costs. Finally, Dr. Meissner discovered that as a rule, companies generating less than $200 million in annual revenue do not make money, whether IDM or fabless, demonstrating that, for small companies at least, the fabless model is no answer to attaining profitability. Merits of consolidationOn the subject of consolidation, Dr. Meissner stated that there are very few IDMs or fabless companies - small/medium or large - that have become successful through consolidation, citing silicon IC company TowerJazz as a good example in the IDM space. According to public filings, TowerJazz lost $100 million on sales of $400 million in its most recently reported year, due largely to overhang from the consolidation process between Tower Semiconductor and Jazz, including depreciation and debt. While TowerJazz might manage to show a reasonable EBITDA on a non GAAP basis, it does not do so by GAAP measures. And, Dr. Meissner added, the folks at TowerJazz are smart, creative and operationally capable, but it is tough to fundamentally solve a profitability problem through acquisition alone. Applying this example to the optical components space, Dr. Meissner said that while he believes consolidation can improve matters to some extent through cost savings if done well, he does not see it as a realistic path to achieving profitability. Development of the fabless modelDr. Meissner then went on to outline the origins and potential advantages of the fabless model, particularly with respect to the optical components industry. He started by noting that the fabless model emerged in the 1980s, primarily as a way to utilise excess capacity in foundries. LSI Logic was cited as an early adopter of the model when it began sending its semiconductor wafer production to TI and Hitachi rather than doing it in-house. A key factor in the early development of the fabless model was Taiwan, Dr. Meissner observed. In Taiwan the government offered subsidies and cheap capital to encourage companies to establish fabs in the country, with TSMC being a good example of one such company that set up there and now makes plenty of money. A fundamental difference between then and now, Dr. Meissner noted, is that in the 1980s there was a need for factory capacity, which allowed companies such as TSMC to establish and grow. Another element was the effect of Moore's law driving below 1 micron, which rapidly pushed up the cost of setting up a fab to around $1 billion. (He also pointed out that the driving forces for increased capacity and Moore's law do not play a role in the optical component space). Together, these factors led to companies building fabs to meet the demand for capacity while others sought to avoid the cost of doing so by moving to a fabless operation. Dr. Meissner went on to describe some of the specifics of establishing a fab. A central element is the lithocell, the most expensive single item in a facility. For a 65-nm process technology plant, the litho tool will cost tens of millions of dollars. In addition the fab needs track and control for the resist, plus inspection systems, elements that form the tool for mapping constraints for the factory. From there all elements upstream on the line deliver excess capacity, which adds cost. Another factor affecting the cost is process wafer size. As the technology scales down it becomes necessary to add capacity by moving to larger wafers and the cost of equipment enabling conversion to a larger wafer size is significantly higher than that it replaces. Dr. Meissner went on to cite volume as a further factor affecting the optical components sector, "A $100 million company is likely to offer many products, each with annual run rates of 100,000 to 200,000 or more and ASPs of at least $100. Such numbers simply do not apply to our industry. However, Santur may ship only in the order of 100,000 devices at ASPs that are 5-10x higher, and that is considered a very high volume product for the sector". It was noted that a speaker from Heinrich Hertz Institute put annual silicon chip output at around 7 billion inches per year, in comparison to which InP chip output is almost negligible. Dr. Meissner concluded that the availability of fab capacity allows companies to be fabless and to thereby focus on factors such as software and design optimisation and product performance, as well as lowering the barriers to entering the market. It also allows IDMs to multi-source in a way acceptable to their customers and without losing control of their IP. Differentiate for profitabilityAccepting that profitability does not derive from being fabless or through consolidation, Dr. Meissner then posed the question - where does it come from? The answer, he believes, is differentiation. Illustrating the point, Dr. Meissner compared Maxim with the average large IDM (excluding Intel), and Altera with the average large fabless company. Both of these companies show what would generally be regarded as excellent rates of return, he said, while the groups, overall, deliver no more than reasonable rates of return. According to Dr. Meissner, both Maxim and Altera achieve better than average performance through differentiation. It was noted that Maxim has high R&D and engineering costs because of its application-intensive partnership model with customers. Maxim has built a policy of only developing products with a market - and that will allow it to deliver differentiation and make money. Similarly, Altera has high SG&A costs due to maintaining numerous design centres for its many customers, although its R&D costs are relatively low due to off-shoring development facilities for non-core elements, even though it continues to develop core elements in the U.S. "The key is to find differentiation and then engineer the cost structure of the company to the lowest level that still allows that differentiation to be maintained. The business model as regards being fabless of fabed is largely irrelevant to achieving this". This then raises the question of whether the fabless model has any value? Dr. Meissner said he believes the most valuable feature of the fabless model to be the ability to cut time to market for products.  |
|||||||||
