machine learning

Every day it seems like a new portable voice-first device is coming to market. From smart speakers small enough to fit in your pocket to tiny wireless earbuds and voice-activated TV remote controls, we are using voice increasingly to play music, select TV shows, turn on the lights or interact with our smart thermostat. While the popularity of voice-first interfaces has spawned massive diversity in device type, as long as these devices are portable, they have one thing in common: They’re battery-powered, and that could be a problem for consumers who are tired of frequently recharging or replacing batteries. Change the Architecture, Reduce the PowerThe issue lies in the traditional hardware architectures of today’s voice-first devices, which are notoriously inefficient when it comes to power consumption. Such devices rely on a “digitize-first” model of processing voice data in which the heaviest power-consumers, like the analog-to-digital converter (ADC) and the digital signal processor (DSP), do all the heavy lifting up front, right at the start of the audio signal chain. They continuously digitize and analyze 100% of the ambient sound data as they search for a wake word, even if speech is not present and the only sound is noise. Because voice is spoken randomly and sporadically, that continuous digitization of sound wastes up to 90% of battery power.To tackle the battery drain in portable voice-first devices, we need look no further than the human brain. Our brain processes sound very efficiently. Imagine that you are outside your house having a conversation with your neighbor. You are able to focus on what your neighbor is saying because your brain can differentiate between sounds that it should send to the deeper brain for speech processing and sounds that it shouldn’t bother processing further (e.g., dog barks, sirens or car traffic). The brain spends minimal energy up front to decide whether it should spend additional energy on processing down the line. In other words, it saves the most power-intensive processing only for the important sounds.We can mimic the brain’s approach to signal processing by enabling a new “analyze-first” architecture for voice-first devices. This analyze-first approach requires ultra-low-power analog processing technology that can differentiate voice from noise before the sound data is digitized. This keeps the higher-power capabilities in a voice-first system, such as the wake-word engine, in a low-power mode when just noise is present. This approach only wakes up the higher-power chips in the system, e.g., the DSP or ADC, when it detects speech. Like our brain, a voice-first system uses an analyze-first architecture to conserve energy most of the time, saving the heavy lifting, i.e., the wake-word listening, for times when speech is present. The analyze-first architectural approach to always-on listening analyzes the analog microphone prior to digitization, saving considerable power in portable voice-first devices that run on battery. This architectural shift to analyze-first is well worth the investment because it reduces the system’s power consumption in a battery-powered voice-first device by up to 10x. That’s the difference between a portable smart speaker that runs for a month on battery instead of a week or smart earbuds that last for a whole day instead of a few hours on a single charge. Longer battery life in portable voice-first devices generates more good will among consumers, creating another key differentiator for manufacturers engaged in the ultra-competitive race for more users.For more information on the analyze-first architectural approach to voice-first devices, please view our video.Tom Doyle is CEO and founder of Aspinity. He brings over 30 years of experience in operational excellence and executive leadership in analog and mixed-signal semiconductor technology to Aspinity. Prior to Aspinity, Tom was group director of Cadence Design Systems’ analog and mixed-signal IC business unit, where he managed the deployment of the company’s technology to the world’s foremost semiconductor companies. Previously, Tom was founder and president of the analog/mixed-signal software firm, Paragon IC solutions, where he was responsible for all operational facets of the company including sales and marketing, global partners/distributors, and engineering teams in the US and Asia. Tom holds a B.S. in Electrical Engineering from West Virginia University and an MBA from California State University, Long Beach. For more information, visit www.aspinity.com. Aspinity is a member of SEMI-MEMS Sensors Industry Group, which connects the MEMS and sensors supply network, allowing members to address common industry challenges and explore new markets.

New system-on-chip (SoC) devices are driving new memory architectures and photonic interfaces, while specialized new intellectual property (IP) requires analysis down to the nanometer and atomic levels because of single nanometer process nodes. According to Babak Taheri, CTO and EVP of products at Silvaco, a leading EDA Software, semiconductor IP company, a member of SEMI and the ESD Alliance, a SEMI Strategic Association Partner, design technology co-optimization and proven IP are required for this analysis.Taheri recently discussed atoms to systems in next-generation SoC designs with Nanette Collins ahead of ES Design West, co-located with SEMICON West, July 9-11 at the Moscone Center in San Francisco.ESD Alliance: For years now, the assumption is that each new chip design is more complex than the last. Why are the latest SoC designs even more complex than before?Taheri: New SoC devices for mobile phones, automobiles, intelligent edge nodes, big data compute and storage are adopting artificial intelligence and machine learning technologies. This is driving new compute, data flow, as well as memory architectures that are bandwidth-limited and some require photonic interfaces.One common denominator in present SoC design are the numerous blocks of IP. On average, over 85% of these blocks are reused. It’s cost-prohibitive to make these chips over and over again with new IP. According to some estimates, 90% of IP used in an SoC design by 2025 will be reused – only 10% is new technologies. That 10% is significant.ESD Alliance: How so?Taheri: Complex new technologies including flash memory, other advanced non-volatile memory technologies such as MRAM, RRAM and SoCs such as NVIDIA’s Xavier and Apple’s A12 use and reuse design IP at the architectural level.New technologies mean new materials and new processes. Single nanometer process nodes require specialized new IP that needs to be simulated and analyzed down to the nanometer and atomic levels.ESD Alliance: Does the atomic level changes the design equation?Taheri: Yes, it does. Designers need to be able to simulate at the atomic level and understand properties of these materials, and how they behave in at-process and at-device levels. They need be able to simulate the material's nanometer geometries, how molecules behave and how they interact for device operations. When they put together a process and a device, they need to know how the pieces behave and simulate before production.In other words, they run quite a few design experiments and quite a bit of simulation before they finalize the circuits and devices to silicon to save money.ESD Alliance: It’s obvious design automation will continue to have a vital role in design.Taheri: Yes, absolutely. Design technology co-optimization (DTCO) using TCAD solutions and proven design IP are needed to address the span from architecture to device and process physics. The importance of simulation, emulation and design technology co-optimization, along with fully verified and proven IP for SoC design, cannot be overstated. As designers generate devices and processors, they take that up to circuit-level simulation and high-level simulation, schematic capture, extractions and back annotation. They can go from atoms to simulating systems to the ability to do that under the same umbrella in order to get better chips, better yield and lower cost.Taheri’s talk Next Generation of SoC Design: From Atoms to Systems will be part of the Meet the Experts More than Moore session Tuesday, July 9, at 11:30 a.m. at the ES Design West SMART Design Pavilion. SEMICON West attendees are invited to Moscone Center’s South Hall to learn more about electronic system and semiconductor design and its links to the electronic product manufacturing and supply chain. Register for ES Design West or SEMICON West.Babak Taheri is Silvaco’s CTO and EVP of products, has more than 25 years of design experience. His current role managing Silvaco’s Technology CAD (TCAD), electronic design automation (EDA) and IP product divisions makes him an expert on what’s needed for the design of next-generation system-on-chips (SoCs). Previously, he was the CEO and president of IBT working with investors, private equity firms, and startups on M A, technology and business diligence. Babak received his Ph.D. in biomedical engineering from the University of California Davis with Bachelor of Science degrees in Electrical Engineering and Computer Science and Neurosciences. He has published more than 20 articles and holds 28 issued patents.Nanette Collins is a public relations representative for the Electronic System Design Alliance.

MedTech, autonomous driving and other disruptive technologies will be in focus at the SEMI Industry Strategy Symposium (ISS Europe), 31 March - 2 April 2019 in Milan, Italy, as top European executives, researchers and academics gather to explore solutions to the region’s most pressing strategic, economic and social challenges. Ahead of ISS Europe, SEMI spoke with Mark Purdy, managing director and chief economist at Accenture Research, about Accenture’s Business Futures – four different future worlds set in 2025 based on the collision of trends across demographics, geopolitics, technology, and economics – and what these futures will mean for markets, workforces, operating models and industry value chains. SEMI: At ISS Europe in Milan, you will kick off the symposium highlighting market opportunities of the digital economy and how companies must adapt to competitive challenges. What inspired Accenture’s Business Futures four world scenarios?Purdy: The impetus for our Business Futures really stemmed from a certain dissatisfaction with current approaches to thinking about the future. We were struck by the following puzzle. First, there is no shortage of techniques for looking at the future, from forecasting to trends analysis to conventional scenarios. Second, most decision-makers have more or less the same access to information on global trends. Yet, time and again, we hear stories of businesses going bust or facing major challenges precisely because they failed to anticipate major changes in their industry.The paradox is that we have so much information, but so little real understanding of how the future actually unfolds. So that set us thinking about how to develop a new approach, based on a combination of detailed trend analysis, expert input and creative storytelling – which is what we call “Business Futures.” SEMI: Of demographics, geopolitics, technology, and economics, which trend do you see as particularly critical?Purdy: Actually, the essence of our Business Futures thinking is that it is the collision or combination of different trends – across economics, technology, demography, etc. – that shapes future outcomes, rather than individual trends per se. To a certain extent we tend to become fixated on specific trends and this can lead us astray or cause bad decision-making. For example, in the early 2000s many people saw very favorable trends in the U.S. economy – strong capital inflows, rapidly rising consumer spending, surging stock markets, and rising home ownership rates. Each trend in isolation looked strong and sustainable. But we failed to see how the combination of these trends was fueling risky financial innovation that would eventually lead to the financial crisis and great recession.Technology of course is a key trend. We are seeing tremendous advances in next-wave technologies such as robotics, machine learning, intelligent objects, 5G and virtualization. But we can only truly understand the impact of the technologies – and the business opportunities and challenges they create – by viewing them against a wider backdrop of changes in society, demography, geopolitics and economics. That is what Business Futures strives to do.SEMI: What will these different futures mean for markets, workforce, operating models and industry value chains?Purdy: There will be profound changes in how we think about all of these areas. Markets will become much more personalized and interactive. Technology will be increasingly integrated with humans, fueling innovation in areas such as personalized healthcare and preventative medicine. Our notions of distance and capacity will be upended, as new virtualized services enable new ways of reaching underserved customers. Consumers will become increasingly involved in the creation and design of products and services. New methods of innovation, powered by AI and virtualization, will come to the fore. New entrants will come from unexpected quarters, enabled by new technology. The upshot will be massive disruption and disintermediation of value chains across many sectors.SEMI: What can Europe do to prepare?Purdy: There are no simple answers, and the correct course will vary by country, but there are some basic things to get right. First, different countries need to understand their comparative advantage – for example, whether it is in services, new technologies, advanced manufacturing or resources – and work with the grain of these different futures. Second, countries need to ensure that they have the basic conditions – regulation, organizational adaptability, workforce flexibility, skills, and innovation infrastructure – to capitalize on the productive potential of new technologies such as AI, virtual reality, and the Internet of Things (IoT). Third, we need to create educational systems and workforce learning methods that emphasize creativity, problem solving and innovation – precisely the skills that will be most needed in an age of intelligent machines. SEMI: What are your expectations for the summit in Milan and for the future?Purdy: I’m very much looking forward to the ISS Europe Summit in Milan. As an economist, I believe we are at a pivotal moment in the semi-conductor industry, driven by waves of technological change and rising geopolitical frictions and uncertainty. With so many industry leaders and experts coming together at the Summit, I’m confident that our discussions will help point a way forward!Mark Purdy is managing director of economic research at Accenture Research. His research examines issues at the intersection of economics, technology and business. He has published widely in tier-1 media and specialised publications on topics such as China’s economy, emerging-market geographic strategy, inclusive economic growth, business futures and the economic impact of new technologies such as the Internet of Things and artificial intelligence. A graduate of Trinity College Dublin, he speaks on these topics at conferences and seminars around the world.Serena Brischetto is a marketing and communications manager at SEMI Europe.

For nearly two decades, Sean Ding, CTO and chief scientist of Alibaba Cloud IoT, has worked in software and algorithm architectures, sensing, semiconductors, systems and cloud computing – all areas that have contributed to the rise of the Internet of Things (IoT). It’s no surprise, then, that Alibaba is leading next-generation innovation for the IoT. Ding will bring his expertise to his role as moderator of Brave New World - MSIG Conference on AI+IoT 2019, a half-day forum March 20, 2019, at SEMICON China in Shanghai, China. Maria Vetrano of SEMI spoke with Ding about technologies key to the IoT era including MEMS, sensors, artificial intelligence (AI), edge gateways and cloud computing. SEMI: MEMS sensors are widely used in IoT devices. What is the relationship between AI and MEMS sensors?DING: While MEMS sensors and AI will increasingly co-reside in end-user devices, I do not recommend adding AI next to the sensor (in the same package). That’s because designers continue to use the ASIC for signal conditioning, so A/D converters are still required. Rather, we should look to edge gateways to carry the majority of the workload, including deep learning, because this reduces system complexity and power consumption.SEMI: Why are smarter sensors shifting data processing and analytics to the edge of IoT devices?DING: Data processing and analytics are very important for IoT devices, but we need to focus on understanding the data, parameter calibration and more. The MEMS sensor industry should leave big data analytics to edge computing and cloud computing because AI requires deep learning, demanding a huge amount of data.The challenge is to find the sweet spot for data processing right next to the sensor element.SEMI: What is China’s evolving role in innovation in MEMS sensors for IoT devices?DING: At present, the MEMS community in China needs to figure out how to innovate instead of copying existing technologies, a low-margin business that will not help to grow the industry. One reason why I am so pleased to see the MSIG Conference on AI+IoT in China is that it will encourage greater creativity in the MEMS community in China, and this will ultimately lead to Chinese companies and R D institutions leading innovation rather than copying it.SEMI: What is the right approach to combining smart MEMS sensors with AI in IoT devices? Why is this important for both domestic Chinese and international markets?DING: Combining data from sensors with cloud-edge computing is the right approach. As sensor companies increasingly provide end-to-end solutions, such as “sensor+ firmware + SaaS + app,” we will realize easier and faster integration of sensors in IoT applications.This is incredibly important because China today is the world’s biggest market for IoT hardware. China has 2,000-plus design houses, 200-plus OEMs and thousands of distributors. That said, we still see a highly fragmented market that will benefit from a faster integration methodology.Faster integration of MEMS sensors and AI/machine learning for IoT hardware will benefit designers in international markets as well.SEMI: What do you hope MISG Conference on AI+IoT attendees will take away from the forum? DING: MEMS sensors are highly fragmented, reflecting the highly fragmented applications in which they play. The MEMS sensors industry should figure out how to provide one-stop-shopping solutions for vertical markets. This will speed the scalability of applications and expedite the growth of sensor production. Sean Ding (柯镇) will moderate Brave New World - MSIG Conference on AI+IoT 2019 at SEMICON China on Wednesday, March 20, 2019, at Kerry Hotel Pudong in Shanghai, China.This conference has been organized by the MEMS Sensors Industry Group (MSIG). Register today to connect with Sean Ding and featured speakers at the event.Speakers at the MSIG Conference on AI+IoT 2019 at SEMICON China include: Welcome and Introduction / 欢迎辞Carmelo Sansone, Director, MEMS Sensors Industry Group (MSIG), a SEMI technology community AI Needs Accurate Data – MEMS Sensors Can Provide It / MEMS传感器为人工智能提供真实数据Andrea Onetti, Group VP of Analog MEMS Group, GM of MEMS Sensor Division, STMicroelectronics Enhanced IoT Edge by Smart Sensors / 智能传感器助力IoT边缘智Bennini Fouad, Regional President Asia Pacific, Bosch Sensortec Horizon AI Processor Solution, Enable Industries in AI Time / 地平线AI芯片解决方案，赋能千万业Carl Zhang 张永谦, General Manager/VP, Smart Chip Solutions Division, Horizon Robotics Inertial Sensors in AI Applications / 运动传感器AI应用案例Ben Lee 李彬 , CEO, mCube Ultra-Low-Power Solutions: an Ecosystem Approach / 超低功耗的生态链解决方案Carlos Mazure, IEEE Fellow, Chairman Executive Director, SOI Industry Consortium High-Integrity, Fault-Tolerant Open Inertial Measurement Platform for AI-based Vehicle Automation / 适用于人工智能车辆自动控制的高集成及容错的惯性测量开放平台Dan Dempsey, Senior Director of Automotive, ACEINNA Maria Vetrano is a public relations consultant at SEMI.

Constant coverage of an invigorating topic like machine intelligence in the media often urges us to consider its use in EDA technology. As is often the case, there are many myths and falsehoods that consume our time and effort when trying to apply machine intelligence to EDA. This article aims to uncover the myths and to provide helpful advice on applying machine intelligence to your EDA project or product.Value PropositionFirst, there needs to be a clear value proposition for adding machine intelligence to an EDA product. Using machine intelligence to create a me-too product adds no value. EDA customers are too busy to understand or care about an EDA tool’s underlying technology. They just want to use the tool and get results. If the tool delivers value, if it delivers tangible benefits, then they’ll use it. Otherwise, they won’t.Currently, EDA tool developers are already experimenting with AI and machine intelligence without considering this fundamental truth – without a higher-end objective. AI must deliver something better or new, whether a speed advantage, a performance advantage, new features, new insights, or perhaps even something pleasantly surprising. Before you write a single line of AI-enhanced code, you need to clearly understand how AI will enhance the product. What is the value proposition?Use ModelThere’s a major barrier to customer adoption of AI and machine intelligence technology for EDA tools: EDA users are averse to make decisions based on probabilistic results. Instead, half a century of EDA tool use has conditioned them to expect deterministic outcomes from their tools.Back in 2003, a prominent visionary and EDA investor was quoted in an interview, saying: “If I open my eyes five years from now, all static analysis in VLSI will be statistical.” Many EDA luminaries have been proven wrong over time for betting that EDA users will accept statistical results. As enthusiastic as I am about using machine intelligence to improve EDA tools, I must urge caution based on the history of EDA failures that employed a probabilistic use model. Decision-makers and EDA tool users want to see deterministic answers to questions about yield or slack, not probabilistic ones.Our experiences at Paripath in developing the PASER (Paripath Accelerated Simulation Environment) tool also bear this out. We discovered that delivering results 50x faster but with 92% accuracy was simply not good enough for end users. EDA users only started to use PASER when its answers became 98+% accurate. To be adopted in the production flow, the tool had to deliver 99% accuracy.Data EngineeringThere are specific ways to achieve these accuracy goals. The first is data engineering. Machine intelligence is a new approach to EDA tool development and it needs to be trained on a data set. If the data is poor or incomplete, training will create an inaccurate model. Fundamental software-development rules still apply. Garbage in, garbage out.Without good training data, there’s no way for you to build good neural-network models. If you train a model with garbage data, you’ll get a garbage model. You must cleanse the data before you use it for training. Otherwise, the model will draw inaccurate conclusions and customers will not use your tool. The model is not to blame here. The model’s not wrong. The problem lies in poor data engineering, poor data cleansing, and a lack of discipline to prepare input data.High DimensionalityNext, machine intelligence has a unique ability to quickly solve problems of high dimensionality. Pure EDA problems often have high dimensionality. Over the years, EDA developers have perfected the art of segmenting the problems into sequencing solutions with lower dimension. Machine intelligence technology can handle problems with thousands of dimensions, but you need to be careful when tackling problems that have high dimensionality. Too many dimensions can produce confused or inaccurate results with AI and deep-learning technology.It helps to visualize the problem and to analyze the data set before using the data to train an AI-enhanced EDA tool. Several visualization methods can help. For example, t-SNE (t-Distributed Stochastic Neighbor Embedding) lets you reduce a data set’s dimensionality from a very large number to a much lower number. Figure 1 shows a high-dimension dataset with a dimensionality of 2000, which has been reduced to a low dimensionality of 3. Figure 1: Visualizing the Data Set with Lower Dimensionality Reducing the dimensionality of a data set to 3 using t-SNE and visualization allows you to quickly see whether the data set defines an easy or a difficult problem. If the problem is difficult, you’ll likely need to lower the problem’s and the data set’s dimensionality before using the data to train a neural network.Technology SelectionOne factor that determines whether it will be easy or difficult to incorporate machine intelligence into your EDA tool is your choice of AI development tools. AI researchers have developed a long list of frameworks, libraries, and languages that they use to develop AI and machine-learning software. Frameworks and libraries such as TensorFlow, Caffe and MXNet are most popular for developing deep-learning models.However, these tools are not yet popular with the EDA development community. The languages of choice in the EDA community are traditionally C and C++ for development and Tcl for prototyping and creating user interfaces. The rest of the software world has moved on to newer development languages such as Python, Java, R, and such. Moreover, machine-learning development segments into two distinct processes: training (i.e. generating the model) and inference (i.e. using the model).Another question to consider is where to generate the model – at the vendor site or the customer site?Consequently, fitting AI and deep-learning development into EDA development environments can feel like fitting a square peg into a round hole. You may need to create corners in your hole.EDA is a very small player in the overall software market. Relatively few software developers are familiar with writing EDA tools. It’s best to select AI and deep-learning development tools that can provide some sort of interface that’s compatible with EDA’s development tools of choice. Some AI frameworks have lower-level C and C++ interface layers that provide a familiar entry point for experienced EDA developers.At Paripath, we chose TensorFlow for exactly this reason. TensorFlow has a lower-level C/C++ interface. Although the resulting development path becomes a longer one using this approach, it’s a more familiar path for EDA developers and therefore it’s a path that can ultimately lead your EDA development team to success. An elaborate study of comparing these frameworks has been published in the book Machine Intelligence in Design Automation.Integration into Legacy SystemsWhen you understand the value that you expect machine intelligence to add to your new EDA tool, when you’ve cleansed and then analyzed the data set, and when you have selected an appropriate set of development tools, you’re finally ready to add machine intelligence to your EDA development. There are two use models for AI-enhanced EDA tools. The first uses a trained model to guide the EDA tool’s decision-making. In this use case, the trained neural network doesn’t change. The software’s accuracy doesn’t improve with use unless the company that developed the EDA tool retrains the underlying neural network. This use case follows the familiar, existing use case associated with EDA tools developed using deterministic algorithms.For the second use case, the end user is able to retrain the underlying neural network, which allows the EDA tool to produce better, more accurate results over time. This use case produces a win/win situation because end users are able to hone their tools and improve them over time, without help from the EDA tool vendor’s application engineers. If the retrained models are also sent back to the EDA developer for incorporation into newer versions of the tool, all users benefit from other users’ training data.It’s not clear how you’d support this second use case in the current EDA business environment where most data sets are proprietary and are carefully guarded. Most large EDA tool customers want to keep their data in house under tight control. Even with this somewhat restrictive situation, however, EDA tools benefit from the incorporation of machine intelligence because each EDA tool customer can customize the tool and improve its results.Machine intelligence has much to add to EDA tools’ capabilities. Only time will tell if the customers want and will accept these new capabilities. Rohit Sharma, founder and CEO of Paripath Inc., is an engineer, author and entrepreneur. He has published many papers in international conferences and journals. He has contributed to electronic design automation domain for over 20 years learning, improvising and designing solutions. He is passionate about many technical topics including machine learning, analysis, characterization, and modeling. It led him to architect guna - an advanced characterization software for modern nodes. Sharma has written a book titled “Machine Intelligence for Design Automation.” You can download code examples and other information here.Note from SEMI-ESD Alliance: ESD Alliance’s Interoperability Committee brings together the industry to discuss interoperability. By focusing the efforts of the electronic system design community onto key compute operating systems, the Interoperability Committee seeks to define a stable, interoperable environment for tools and streamline the resources required to support these environments. The EDA Industry OS Roadmap presents guidelines to EDA vendors and customers for compute platforms to target for design starts. Learn more and view the OS Roadmap overview at our website.

Semiconductor fabs have been getting smarter and smarter over the past 30 years. It’s a natural evolution – the direct outcome of numerous continuous-improvement efforts. The really important difference on the road to smarter fabs, the one change that’s enabling the Industry 4.0 revolution, is the concept of a cyber-physical system or digital twin. If you don’t have a thorough, detailed, high-fidelity digital twin of your entire fab operation, then you cannot have “Smart Manufacturing.” That’s really the definition of a smart site. A digital twin is simply a requirement for all smart factories of the future. One caveat: No matter what you build today from a smart perspective, your digital twin’s fidelity will improve over the next 20 years. A factory’s digital twin has two facets: the operational aspect and the yield aspect. Each of these two facets places different requirements on a database including the types of data, the frequency of data generated, the retention of data, and even the AI/ML techniques used to analyze the data. A combination of these data requirements are needed to create a digital twin – the virtual representation of your entire factory operation, whether it’s on the wafer-fab front end or the assembly and test back end. What’s most important here is that facility-wide data sets and databases must be able to communicate with each other using refined summary statistics to create a practical digital twin. For example, a lot of information is collected on the yield side to feed the deep-learning models needed to manage processes. However, the factory scheduler, driven largely by the smart operational database, needs only summary statistics from the yield database to be able to act in the next moment or over the next 24 hours. Figure 1 illustrates the needs of and the interaction between a smart operational and a yield database. Figure 1: The Operational and Yield databases in a Smart Factory need to exchange summary statistics. Today, we find that although these databases generally speak to each other in smart factories, they’re still not sufficiently connected to permit the use and analysis of data needed to realize the full potential of a smart factory. That level of interconnectedness is still in the future. Some solution providers have created what is essentially a “smart learning warehouse” (“database” has become too limited a term here). This warehouse collects, analyzes and learns from the extensive amount of information that a fab generates. Game-changing, more holistic applications become possible when this information can be combined in new and informative ways. As it turns out, a data source is just a data source, but users in different factory areas need to extract different information from these common data sources. They need different applications and portals – in other words “views” – that are adapted and adjusted for each area’s needs. Aren’t we smart enough? Some people think that 300mm fabs are already smart. That’s true. They are. But, they could be a lot smarter. No 300mm fab in use today has attained the full, utopian vision of what a smart factory can deliver over the next 10 years. When you finally integrate all of the disparate databases in a fab – when you’re able to use all of those different data sources as one common data source – that’s when your Smart Factory will have the ability to self-optimize its future actions and react quickly to real-time events. The largest semiconductor manufacturers tend to develop these smart factory applications on their own. The remaining semiconductor fabs need to work together with other fabs and their solution providers to develop these smart factory applications. Why now? Why is everyone talking about “Smart” now? It’s because the semiconductor industry has helped to create all of the enabling technology: the compute power, the networking and networking standards, and even the industry’s maturation into a multi-tiered organization of solution providers. We’ve reached the point where we can collect data from a widespread sensor network along with tool-health data and we can then warehouse this data so that it can be applied to more intelligent decision-making. While there may be one or two sensors on a tool today, in the future there will be many such sensors connected over an IoT network or networks that provide mountains of data to the warehouse. All of this data will feed into the digital-twin version of the fab. One of the biggest changes on the horizon made possible by all of this accessible data is advanced scheduling. Despite all of the automation advancements made over the past 25 years, including robotic handling, it’s still hard to decide “where, what, and when?” for every single lot in the factory. Today, no factory in the world is more complex than a semiconductor fab. Optimizing a semiconductor manufacturing process is the most complex manufacturing-optimization task in the world. Do it for ROI ROI is the chief reason for having a digital twin. Once you can make a truly smart, holistic schedule of the fabs operations — not a dispatch or rule-based dispatch list — then you can create an operationally smart factory. Rule-based dispatching systems primarily focus on tools and tool-centric views. Although they incorporate knowledge from current WIP and tool conditions to make decisions better than simple dispatch systems, smart factories are not just about tools and the current WIP at them. Smart factories use the status of every tool and lot in the factory to make fab-centric optimizations instead of tool-specific optimizations. Once you have a digital twin, you’re optimizing for global functions such as line linearity and on-time delivery. These functions are not just about the moment. The transition to a smart factory thus represents a huge philosophical change. When you know exactly what’s going to happen in a factory over the next 12 hours for every single lot, every single wafer carrier, and every single entrance port of every tool in the factory, then you suddenly have control over the factory’s idle time. You know when you can optimally perform PM (preventive maintenance). You know how to best redirect material or labor resources to maximize output. You can create a smart schedule for every maintenance person in the factory that comprehends each person’s skill set and tool downtime so that there’s no negative impact on the factory’s productivity. You can only do all of this when you know the future. Figure 2 illustrates the opportunity. Imagine that a factory contains 1,500 tools. Use of these tools is scheduled for the next twelve hours. The information depicted in Figure 2 encompasses process changes from one chemistry to another, implant changes, reticle changes, and the status of every single consumable for all 1,500 tools. The white spaces that appear between processes in Figure 2 represent opportunities to intelligently schedule events such as maintenance to maximize factory productivity. Figure 2: Smart scheduling permits factory-wide optimization to maximize productivity. Once you have a schedule, you need to translate that schedule into actions or movement. It’s not easy to do this and most material-control systems today make overly simplistic decisions based on modeled assumptions and typical cases rather than the actual time each lot needs to be at a precise location, which can only come from a schedule. Once the data from all of the tools is connected, a smart scheduling system can use the digital twin to make far better process decisions. The larger the factory (or more complex the factory), the more important it is to make smarter decisions. Note: SEMI has a Smart Manufacturing Technology Community. For more information or to get involved, click here. If you would like to discuss Smart Manufacturing more with John directly, he can be contacted at [email protected]. John Behnke is general manager of the Final Phase Systems product line at INFICON.

Nicolas Sauvage, senior director of Ecosystem at TDK InvenSense, will present at the fast-approaching MEMS Sensors Executive Congress on October 29-30, 2018 in Napa, Calif. SEMI’s Nishita Rao spoke with Sauvage to offer MSEC attendees advance insights on Sauvage’s feature presentation.SEMI: What is “autonomy value” and why is it important?Sauvage: How do you increase the perceived value of an electronic device? If it’s an autonomous car, its value is closely tied to the autonomy level — i.e., the independence — that it offers people. Higher autonomy value for a self-driving car, for example, means that even a blind person could use it. It’s been almost two years since Waymo demonstrated this, and here’s the video that shows it.Countless other sensor-based electronic products have their own “autonomy value.” Imagine the need to get medicine to people during a humanitarian health crisis. Drones could be your best option because they can deliver to inaccessible or remote locations. Unlike older drones, which require active piloting by a person, a drone with higher autonomy value could deliver medicine to Doctors Without Borders without ongoing human intervention.This drone could navigate objects, such as trees and birds, and would have excellent location-awareness. It could fly through any landscape in bright sunlight or during the night. To increase the drone’s autonomy value, you would need better sensors, including those sensors that can enable sensing in sunny conditions or in pitch-black night, as well as better machine learning.SEMI: In this example, what types of sensors would the drone manufacturer need?Sauvage: The manufacturer would need a “surrounding-sensing” solution that includes ultrasonic and pressure sensors as well as image sensors. Start with high-quality image sensors combined with ultrasonic range-finding sensors — high-accuracy devices that function in all lighting conditions and can detect objects of any color. Add motion sensors and a pressure sensor, which would capture the height of the drone to make known the drone’s location in space. The drone would need this combination of sensors, plus smart sensor fusion, because GPS alone cannot avoid obstacles: its signal can be sporadic in certain parts of the world or in certain terrain, making it unreliable.A key attribute of all these sensors would be low power consumption since the drone would run on battery.SEMI: To what extent might autonomy value cause manufacturers to consider multi-vendor solutions?Sauvage: I would like to see it inspire the MEMS and sensors ecosystem to work together, to arrive at multi-vendor solutions that will benefit humanity through greater autonomy value. Whether we’re looking at autonomous cars, drones, robotics or other applications, there are cases where we need to prioritize safety and security over industry competition. SEMI: Where are we today in terms of achieving true autonomy value – and where are we going?Sauvage: The sky is the limit, literally. Machine learning and surrounding-sensing solutions applied to cars, drones and robots will increase autonomy value to the point where we can justifiably call it artificial intelligence.SEMI: What would you like MEMS Sensors Executive Congress attendees to take away from your presentation?Sauvage: I hope that attendees will recognize the value of ecosystem solutions in increasing autonomy value. Together we can expand the variety of sensor types that address novel use-cases and jobs-to-be-done. Instead of waiting for customers to ask for ecosystem-level solutions, we need to articulate a complete MEMS and sensors supply-chain ecosystem if we want the Internet of Things (IoT) and Industrial IoT (IIoT) to grow more quickly. As senior director of Ecosystem, Nicholas Sauvage is responsible for all strategic relationships, including Google and Qualcomm, and other HW/SW/System companies. He is also responsible for strategic and market-driven goal-setting of our SensorStudio developer program, and driving select partnerships with SoC sensor hub platforms. Prior to joining InvenSense, Nicolas was part of NXP Software management team, responsible for worldwide sales, as well as for P L and product management of their OEM Business Line. Nicolas is an alumnus of Institut supérieur d’électronique et du numérique, London Business School and INSEAD. Register today to connect with Nicolas Sauvage at the event. You can also connect with him on LinkedIn.Nishita Rao is a marketing manager at SEMI.

Part 2 of this two-part piece examines the potential benefits to be realized by pairing human Subject Matter Experts with smart silicon assistants, and what these new arrangements mean for semiconductor device manufacturing. Part 1 explores best-practice perspectives on collecting and utilizing smart data in industries outside semiconductor manufacturing, one of the important takeaways from the Smart Manufacturing panel discussion at SEMI ASMC 2018. So what does this observation (i.e. the field of medicine, in what seems at first glance a big data environment, is really just clusters and clusters of loose small data connected by the collective neural network of highly trained doctors and their colleagues) mean for semiconductor manufacturing? We think it means we need to apply the same level of intense focus that we already devote to instrumented data collection and analytics in the fab to something more: we need to better capture the vast expertise of our engineering and operational talent in semiconductor manufacturing. We think we need to record what the subject matter experts (SMEs) in the fab see, hear, and think as they investigate yield excursions or machine-down problems. We need to effectively combine product, process, equipment and component subject matter expertise / subject matter experts (SME) with big data analytics to more effectively solve manufacturing problems, be they killer or be they chronic. And we must provide structured methods for incorporating inputs from and active participation of SMEs throughout the data analysis lifecycle, from collection and aggregation, through filtering, feature extraction, analysis and optimization. Some of the challenge will be in just how do we make it easy to gather information from SMEs in real time, while standing in front of equipment in the fab. Internet of Things (Iot) devices are emerging to capture and label images and sounds to enable machine learning algorithms to recognize and help diagnose manufacturing problems based on sight and sound, complementing the instrumented data. But we also need to record the thought processes our human SMEs go through in those investigations – perhaps by the SMEs talking to a smart AI-based conversational assistant who helps make “rounds.” Doing contextual analysis on this added data, combined with the instrumented data, will create the equation Human + Machine = AI (Awesome Insight). Sounds reasonable, right? We think artificial intelligence becomes too artificial if you leave the human out of the equation. AI should be augmented intelligence, where we take the expertise and creativity of the human, and combine it with the rapid computational capabilities of the computer, in order to put problem identification and solutions on steroids. But with the already huge advancements to date in data analytics, cloud, and the emergence of AI, why do improvements in quality, machine utilization, and the implementation of predictive analytics in semiconductor manufacturing seem to be creeping along incrementally, and not appearing as dramatic, step-function improvements? Call it Smart Manufacturing, call it Connected Enterprise, call it Advanced Manufacturing, or Analytics, or Cloud, or the Digital Twin … there are no shortages of terms, philosophies, and technologies available, but why aren’t we seeing their rapid adoption? It could be it’s the downside that comes with needing people. “Good business leaders create a vision, articulate the vision, passionately own the vision, and relentlessly drive it to completion.” Jack Welch. We see from other industries that smart manufacturing conversations originating with the executives of a company thinking to implement smart manufacturing programs lead to vision; however, we also see from other industries, and from our own, that realizing this vision has often been a challenge. Why is that? One reason may be that the people who are personally vested in solutions they implemented in the past, as well as those who follow a pattern of ‘how we’ve always done things’, create, inadvertently or not, persistent internal barriers hindering innovative action. Another may be that engagements with the working engineers and managers charged to be smart manufacturing implementers leads to the pursuit of low-hanging fruit, and cautious investments, that often utilize solutions that ultimately cannot scale and integrate. Not to mention the disadvantage of dealing with the legacy equipment, the legacy networks, the traditional thinking, and the lack of consistency in metrics adding to the confusion. Addressing all these barriers requires an alignment in strategy and execution, along with a plan to support the overall vision, often across the entire enterprise, which is no small matter. And then there are the standards. Having and adhering to standards in control solutions, networks, and data becomes critical in achieving real benefits from smart manufacturing. And data security. One of the other big impediments in the smart manufacturing transformation is data and IP security, another key concern (maybe the most significant) preventing us from moving forward more quickly (e.g. to cloud-based solutions) in our industry. More about that in a follow-up. Achieving synergy across all of manufacturing, from connecting equipment horizontally, through the production system (machines processes), and vertically, through enterprise systems and across production facilities, can only occur if we build standards, security, infrastructure, and human engagement throughout our ecosystem and supply chain. In simple form, the steps to do so include connecting assets, collecting and contextualizing data, and then driving business transformation with actionable insights gained from the data. With impact on every function, and every person, in the enterprise, from equipment operators in the fab through the C-Suite in HQ. Maintenance, Engineering, R D, Operations, Scheduling, IT, Procurement, Finance, HR all contribute, collaborate and benefit. Regardless of the technology, from device level analytics to predictive maintenance and optimization, the people that reside in these disparate groups need to come together with the smart machines to create a common strategy to achieve transformational results. Aligning an enterprise’s goals with its human capital is paramount to success. Therefore, we must challenge our team members and ourselves to work outside our comfort zones, and we need to be forever aware of the need for us to grow with the technology. Smart manufacturing is not necessarily about having fewer people in the fab, but it does suggest having people in the fab, perhaps with different, or upgraded, skill sets, who are even more efficient in their roles as a result of the boost they are getting from Industry 4.0. Fortunately, we now have techniques that let us combine the best, brightest, and latest and greatest analytics with our invaluable SMEs throughout the data analysis lifecycle. We’ll not only be able to deliver higher quality semiconductor manufacturing solutions all in all, but we’ll also be providing methods to more easily distribute, scale, maintain, and continually refine those hard-earned solutions. We expect that subject matter experts will continue to put the “smart” in machine-based smart manufacturing today, and for the foreseeable future. SME contributions are not an option, but, rather, an imperative for ensuring a semiconductor manufacturer’s sustained prosperity, much less its survival. Nancy Greco (IBM Watson), Dave Mayewski (Rockwell Automation), James Moyne (University of Michigan / Applied Materials), and Paul Werbaneth (Intevac, Inc.), along with Julie Jacob (Ernst Young), and Carson Henry (Micron Technology), were members of the SEMI ASMC 2018 panel discussing Industry 4.0 and the Future of Commercial Semiconductor Device Manufacturing. All opinions here are purely our own. Please contact Paul Werbaneth via email at [email protected]. The SEMICON West (July 9-11, 2018, in San Francisco) Smart Manufacturing Pavilion features working production equipment on the floor and three full days of speakers providing insights on building the infrastructure needed to enable AI. Equipment from Bosch Rexroth, Cimetrix, Rudolph Technologies, INFICON, Final Phase Systems, OMRON, DISCO and Edwards Vacuum will showcase cutting-edge smart manufacturing technologies. For information on the SEMI Smart Manufacturing initiative and how to get involved, please click here.