9SEPTEMBER 2023that comes with it. Much of this is driven by the computing demand of more complex applications, such as the rise of generative AI or autonomous driving. For example, depending on the input, the cost per inference of ChatGPT can be 4x to 70x more than a standard Google search. In autonomous vehicles, up to 50 percent of power is used for computation rather than the actual driving in city driving scenarios. This need to process increasingly complex data and calculations, at greater speeds than ever before, is underscoring the urgent need for more energy-efficient devices. The semiconductor industry, therefore, is under immense pressure to reduce energy consumption, carbon footprint and other environmental impacts associated with its products and processes. In addition, semiconductors are expected to enable significant efficiency improvements at the system level because, according to the International Energy Agency (IEA), over the next two to three decades two-thirds of the reduction in carbon emissions are expected to derive from improvements in energy efficiencies (with only one-third coming from Renewable Power Generation). High voltage semiconductor technologies therefore (using Si, SiC and GaN) will be critical to achieving Carbon Net Zero initiatives. Industry leaders such as Power Integrations and several startups have developed or are working on products that will address these industry needs.These challenges are immense, but they present great entrepreneurial opportunities as well, and the good news is that talented engineers and entrepreneurs all over the globe are already developing solutions to address these market needs. The startup space is a dynamic and critical piece of the industry ecosystem, where teams can often be more nimble and innovative than larger companies, unburdened by legacy technologies or business models. Semiconductor startup innovation is well underway in a number of areas, and some of the most promising include: New architectures. Semiconductor startups are developing new architectures that can improve the performance and efficiency of semiconductors. For example, some startups are developing new architectures that can better parallelize computation, while others are developing new architectures that can better exploit the power of AI and ML. While computational needs are increasing, Moore's Law (which says that chip performance will double every 18 months) is slowing and the cost of scaling functions in two dimensions with new process technology nodes is becoming prohibitive. A state-of-the-art (sub 7nm) fab can cost more than $20B, but not all functions on a device need to be on cutting-edge technology nodes. This reality has led to chip functions being partitioned into individual chiplets and then being combined on an interposer. This is called Heterogenous Integration. An example of this is Nvidis's H100 where a GPU is combined with several HBM 3D stacks on a large silicon interposer. Apple has also employed a similar approach on its M2 processor.Eliyan is one example of an emerging company in this space, where they are developing chiplet systems that enable greater chip-to-chip bandwidth and speed. Eliyan's technology helps to solve the need to process more data at high speeds by enabling movement of large bi-directional data feeds, helping to make whole systems ­ from chip-board-rack to network ­ run more efficiently. New application technologies. Semiconductor startups are developing new chip technologies that are optimized for applications with specific needs. For example, some startups are developing semiconductors for use in medical devices, while others are developing semiconductors for use in self-driving vehicles where the specs must meet very rigorous standards for cost, weight, power consumption, safety, and more. Recogni is a startup which has developed an industry-leading perception system for autonomous vehicles, enabling real-time object recognition for self-driving vehicles with object detection accuracy up to 1000m at 10-20X more power efficiency than competing solutions. New materials and processes. Advanced processes are being developed and implemented by the leading semiconductor fabricators. These new process nodes are needed to reduce size and improve the performance, efficiency, and cost of next generation semiconductors. Examples of these include advancements in atomic scale processes such as Atomic Layer Deposition and Etch for features such as Gate-All-Around (GAA), FETs, and 3D NAND stacks with very high aspect ratios. Advanced process nodes will require new materials that can withstand higher temperatures and allow for faster transistor switching, high-K gate dielectrics, DRAM cell dielectric films, stacked films for 3D NAND, masks, and other applications.These advancements will require new process equipment and advanced process monitoring tools. Atonarp is one example of a company working in this space, having developed an in-situ metrology platform that enables quantitative, rapid gas analysis required in advanced semiconductor manufacturing processes such as ALD and Etch, helping to greatly improve fab yields. The Atonarp ASTON mass spectrometry platform is also being used for abatement.While the semiconductor industry is undergoing a period of rapid change, this is not the first time. If we look back to even just 5 or 10 years ago versus today, the speed of advancements to meet new market demands has been remarkable. Continued innovation remains the best answer to meeting the challenges we're confronting ­ and the emerging next generation of startups will play a central role in bringing this innovation to life.
< Page 8 | Page 10 >