July 7, 2014, IMEC Technology Forum, San Francisco—A panel considered the likelihood of an inflection point in the semiconductor industry. Debra Vogler from Instant Insights moderated the panel. Panelists were: John Chen from nvidia, Richard Gottscho from LAM Research, Terry Brewer from Brewer Science, Christophe Fouquet from ASML, and Luc Van den Hove from IMEC.
Van den Hove noted the keys to increased collaboration are in the rising barriers such as increased R & D costs, and the cost of new fabs exceeding $10B by '16. The investment for a new design at the leading edge requires huge volumes to achieve any ROI. As a result, the fabs are consolidating and the equipment suppliers have to merge to maintain economies of scale. The industry is splitting into design and marketing companies, and manufacturing companies as even many IDMs go fab light.
Fouquet suggested a holistic approach that integrates lithography, metrology, and computational lithography can provide wider process windows. These wider windows, coupled with design intent change the possibilities in manufacturing. For example, combining litho, metrology, and computation into a closed loop can open a process window from 0.5nm to over 5nm. This integrated approach improves overlay performance and also reduces cycle times.
Brewer went pedantic on the issue. He questioned whether an inflection is happening or if we are confusing cause and effect. Moore's law is really an observation and is not immutable. The industry has always had cycles and variations, but has not done well in retaining the added value. The semiconductor percentage of the value chain continues to drop over time.
New consumer value is at the chip and higher levels and alternate technologies are coming online. The industry is gearing up for the IoT, but the IoT is not another "next big thing". These new markets are large in aggregate, but small individually, and call for a new business model.
Gottscho suggested that the R & D costs for new materials and integrated flows are increasing at an exponential rate. Consumers are buying mobile and storage devices where cost-power-performance-and attractive packages are important. These trends are reducing the number of suppliers and semiconductor manufacturers. This situation is a significant change from the past, and now the best technology at the lowest cost is the prime target.
The increase in consolidation is driving R & D down to the suppliers, so the horizontal silos may move and companies might become more vertically integrated within their segments. The industry has to take action since scale is no longer the provider of solutions and new functions. the industry needs to increase and improve standards, collaboration, and do these things in an environment that maintains IP protection.
Chen declared that graphics functions are an approximation for pixels and vertices. The graphics community has always looked for more transistor speed at a fixed power on every process node. Unfortunately, the transistor performance didn't increase at the 20 nm node, but a change to FinFETs below 20nm improved performance.
As a proof, the mathematical definition of an inflection point is when the second derivative of the function changes sign. Double patterning is a discontinuity that caused process complexity and its proxy of mask steps to jump, resulting in greatly increased die costs. Unfortunately, the die yields are also inversely proportional to the number of mask steps, so die yields are dropping. We have to get to zero defects to increase the yields and reduce the die costs. The alternative is to move to 450 mm wafers, which are overdue for production based on the past history of wafer size increases.
People and connected devices need to find a way to upgrade?
Gottscho proposed a prescriptive change to allow devices to self-upgrade to change functions and make life easier.
Chen added it depends on consumer demand, but affordability is still an issue for many.
Brewer commented that for computer chips, the performance is good enough, and this level of performance is moving into mobile devices. The model of the next big thing is dying, but people always want something.
Should we try to define technology trends from user requirements?
Fouquet stated that people have many devices and Moore's law drives innovation. Consumer needs are a slower driver than Moore. Much of the needed technology already exists, so we still need something like Moore's law to drive technology, generate apps, and drive demand for new devices.
Van den Hove observed that the drivers of technology push versus app pull are both needed. It is very hard to predict needs, so we need a diversity of knowledge to expand markets.
Brewer added that the OEMs like Apple make much more money than the semiconductor manufacturers.
Is this like airplanes, where the speed of planes continues to increase over time?
Chen responded that Boeing considers that the airlines are not in a plateau, but are still growing. One reason is that speed is not the key factor, but the issue is in keeping flying affordable for as many people as possible. Consumers are interested in new technology until it is good enough. At that point there is no additional value in increasing speed or performance. Companies need to innovate in areas where people are still willing to pay.
Consolidation reduce innovation? Scale?
Gottscho considered that consolidation helps innovation and scale. Niche functions will go to the faster-moving smaller companies, since custom functions are not viable for the big companies. These smaller companies can change chips versus other functions, and can change the requirements to use older technologies.
Van den Hove offered that there are lots of opportunities. For example, medical diagnostics is an area with lots of startup companies. The big companies may move to serve a market at critical mass, so the existing core technologies are still viable.
Chipmakers are not interested in small volumes, but customers are diversifying. Market diversification requires small batches as apps spread?
Van den Hove responded that small lot markets exist and look likely to grow.