This industry comprises establishments primarily engaged in manufacturing semiconductors and other components for electronic applications. Examples of products made by these establishments are capacitors, resistors, microprocessors, bare and loaded printed circuit boards, electron tubes, electronic connectors, and computer modems.
The emerging manufacturing sectors of semiconductors and electric vehicles are being heavily recruited by cities and states. The build out may permanently change supply lines.
The automotive industry has an exponentially growing demand for chips, while chip makers struggle with exponential process data growth
Improving Yield With Machine Learning
Machine learning is becoming increasingly valuable in semiconductor manufacturing, where it is being used to improve yield and throughput.
Synopsys engineers recently found that a decision tree deep learning method can classify 98% of defects and features at 60X faster retraining time than traditional CNNs. The decision tree utilizes 8 CNNs and ResNet to automatically classify 12 defect types with images from SEM and optical tools.
Macronix engineers showed how machine learning can expedite new etch process development in 3D NAND devices. Two parameters are particularly important in optimizing the deep trench slit etch — bottom CD and depth of polysilicon etch recess, also known as the etch stop.
KLA engineers, led by Cheng Hung Wu, optimized the use of a high landing energy e-beam inspection tool to capture defects buried as deep as 6µm in a 96-layer ONON stacked structure following deep trench etch. The e-beam tool can detect defects that optical inspectors cannot, but only if operated with high landing energy to penetrate deep structures. With this process, KLA was looking to develop an automated detection and classification system for deep trench defects.
Finding Frameworks For End-To-End Analytics
New standards, guidelines, and consortium efforts are being developed to remove these barriers to data sharing for analytics purposes. But the amount of work required to make this happen is significant, and it will take time to establish the necessary level of trust across groups that historically have had minimal or no interactions.
For decades, test program engineers have relied upon the STDF file format, which is inadequate for today’s use cases. STDF files cannot dynamically capture adaptive test limits, and they are unable to assist in real-time decisions at the ATE based upon current data and analytically derived models. In fact, most data analytic companies run a software agent on the ATE to extract data for decisions and model building. With ATE software updates, the agent often breaks, requiring the ATE vendor to fix each custom agent on every test platform. Emerging standards, TEMS and RITdb, address these limitations and enable new use cases.
But with a huge amount of data available in manufacturing settings, an API may be the best approach for sharing sensitive data from point of origin to a centralized repository, whether on-premise or in the cloud.
Origin of the FOUP
The Race To Zero Defects In Auto ICs
While semiconductor test engineers are making great strides on isolating fab-generated defects, assembly engineers are quietly focusing attention on improving inspection and processing of equipment data to catch latent defects. This is a big deal for automotive electronics. According to a BMW presentation at the 2017 Automotive Electronics Council reliability workshop, most semiconductor devices fail within the car’s warranty period.
The carmaker noted that 22% of warranty costs are due to electronics and electrical control units. Of those failed parts, BMW said 77% of the failures are semiconductor devices, and 23% of the parts are isolated to active and passive components. Of those semiconductor failures, 48% were due to systematic fails, 24% to test coverage, 15% to random failures, and 6% were retested and did not fail the second time. The failure pareto was also broken down to 41% final test, 24% front-end processing, 22% design, and 12% assembly.
For assembly facilities to deliver 10 dppb quality to their automotive customers, they need to learn from customer returns. This requires investment in assembly equipment data collection and traceability. Latent defects that become activated during the warranty period yet pass electrical test necessitates 100% inspection to screen for these failures. Yet all this investment in more inspection and data collection places a financial strain on traditionally inexpensive assembly operations. There is constructive tension between assembly facilities and their automotive customers, as they are both cost-sensitive. Still, somehow this pathway to 10 dppb will be funded.
“We see AI as a disruptive technology that will in the long run eliminate, and in the near term reduce the need for verification,” says Anupam Bakshi, CEO and founder of Agnisys. “We have had some early successes in using machine learning to read user specifications in natural language and directly convert them into SystemVerilog Assertions (SVA), UVM testbench code, and C/C++ embedded code for test and verification.”
There is nothing worse than spending time and resources to not get the desired result, or for it to take longer than necessary. “In formal, we have multiple engines, different algorithms that are working on solving any given property at any given time,” says Pete Hardee, director for product management at Cadence. “In effect, there is an engine race going on. We track that race and see for each property which engine is working. We use reinforcement learning to set the engine parameters in terms of which engines I’m going to use and how long to run those to get better convergence on the properties that didn’t converge the first time I ran it.”
Deep Learning For Industrial Inspection
Where And When End-To-End Analytics Works
To control a wafer factory operation, engineering teams rely on process equipment and inspection statistical process control (SPC) charts, each representing a single parameter (i.e., univariant-based). With the complexities of some processes the interactions between multiple parameters (i.e., multi-variant) can result in yield excursions. This is when engineers leverage data to make decisions on subsequent fab or metrology steps to improve yield and quality.
“When we look at fab data today, we’re doing that same type of adaptive learning,” McIntyre said. “If I start seeing things that don’t fit my expected behavior, they could still be okay by univariate control, but they don’t fit my model in a multi-variate sense. I’ll work toward understanding that new combination. For instance, in a specific equipment my pump down pressure is high, but my gas flow is low and my chamber is cold, relatively speaking, and all (parameters) individually are in spec. But I’ve never seen that condition before, so I need to determine if this new set of process conditions has an impact. I send that material to my metrology station. Now, if that inline metrology data is smack in the center, I can probably disregard the signal.”
The X-Ray Tech That Reveals Chip Designs
When you’re baking a cake, it’s hard to know when the inside is in the state you want it to be. The same is true—with much higher stakes—for microelectronic chips: How can engineers confirm that what’s inside has truly met the intent of the designers? The new version of our X-ray technique, called ptychographic X-ray laminography, can uncover the interconnect structure of entire chips without damaging them, even down to the smallest structures. Using that technique, we could easily discover a (deliberate) discrepancy between the design file and what was manufactured.
Although we can already tell a lot about an IC from just the layout of its interconnects, with further improvements we should be able to discover everything about it, including the materials it’s made of. For the 16-nm-technology node, that includes copper, aluminum, tungsten, and compounds called silicides. We might even be able to make local measurements of strain in the silicon lattice, which arises from the multilayer manufacturing processes needed to make cutting-edge devices.
Why The World Relies On ASML For Machines That Print Chips
The cost of defect failures is starting to spiral out of control, and the cheapest insurance against this is more Metrology and Inspection. One of the changes the industry is adopting is advanced packaging as the primary driver to increasing semiconductor performance. The push to advanced packaging has an entire set of consequences, namely newer packaging technology and a new vector of failure.
Additionally, Metrology and Inspection spending tends to ramp before the rest of the tools, and that is why they should continue to grow so robustly in 2022 given that large fabs are just starting to come online. Metrology and inspection ramps are likely happening currently for the N3 and N5 nodes at TSMC and Intel.
Semiconductor growth through as-a-service models
This report examines As-a-Service (AaS) as an increasingly relevant competitive growth model for the semiconductor industry. Originally successful in the software realm, AaS business models are poised to help semiconductor companies fuel growth, boost revenue, innovate faster, and deepen relationships with customers. When planned and implemented correctly, AaS can substantially increase shareholder value and improve predictability of revenue. However, if executed poorly, it can negatively impact a company’s bottom line.
Improving PPA In Complex Designs With AI
The goal of chip design always has been to optimize power, performance, and area (PPA), but results can vary greatly even with the best tools and highly experienced engineering teams. AI works best in design when the problem is clearly defined in a way that AI can understand. So an IC designer must first see if there is a problem that can be tied to a system’s ability to adapt to, learn, and generalize knowledge/rules, and then apply these knowledge/rules to an unfamiliar scenario.
The Extreme Engineering of ASML’s EUV Light Source
Big Payback For Combining Different Types Of Fab Data
Collecting and combining diverse data types from different manufacturing processes can play a significant role in improving semiconductor yield, quality, and reliability, but making that happen requires integrating deep domain expertise from various different process steps and sifting through huge volumes of data scattered across a global supply chain.
How Japan Won Lithography (& Why America Lost)
Chip floorplanning with deep reinforcement learning
Inside Intel’s Bold $26 Billion U.S. Plan To Regain Chip Dominance
What’s Harder to Find Than Microchips? The Equipment That Makes Them
We typically associate microchips with the latest and greatest technology, but it turns out that most of the chips that go into the products we use are made with older manufacturing techniques. No one knows precisely what proportion of the world’s microchips is made on used equipment, but Mr. Howe, owner of SDI Fabsurplus, estimates it might be as much as a third.
TSMC is expanding its capacity to make older chips by building a new plant for that purpose in Japan. Intel has no plans to build new capacity for manufacturing older kinds of chips, and continues to concentrate on making bleeding-edge chips, says Lisa Spelman, a vice president in Intel’s data-center group.
The Semiconductor Health and Cancer Problem
How ASML Won Lithography (& Why Japan Lost)
Economics of the FPGA
Successive generations of chips are becoming harder to justify where smaller process geometries become nonlinearly more expensive. An increased integration however, drives up packaging complexity and cost. This increase in complexity grows design time and expense. Because of these factors, the margins with which to gain profit from the volume production of ICs are reducing. First, the increase in competition allows for more consumer choices, which reduces per-product volumes. This increase in competition also reduces the life duration and lifetime volume of products. New compute-intensive nodes and technologies must be increasingly agile, not only to support the changing market demands but also to keep up with upgrading deep-learning models. There is an apparent lack of FPGAs being used for AI applications that sit in the “middle of the road” in terms of complexity, causing designers to rely either on custom chiplets or embedded processors for hardware acceleration. The new FPGA economy paradigm opened by Efinix frees designers to more flexibly innovate in a realm that will deliver revolutionary benefits to society. This once-in-a-lifetime quantum shift in product design possibilities is providing an inflection point away from the dead end of custom silicon and into the customizable blank slate of FPGA fabric.
Moore’s Law Could Ride EUV for 10 More Years
ASML expects that chip makers ramping up production with the new technology initially will use 0.55 NA for a cost-saving single-expose EUV process for advanced wafer layers, while using multi-pattern 0.33 NA along with older lithography technology for more mature nodes. As the single-expose 0.55 NA technology reaches its limits, somewhere around six years from now, ASML predicts that chipmakers will once again resort to multi-patterning to reach even more advanced nodes with higher transistor densities. In the next few years, ASML’s introduction of 0.55 NA tools will help leading semiconductor foundries like TSMC overcome obstacles they are now encountering at the 3nm chip process technology node.
The Dutch company is the world’s lone supplier of EUV equipment. In 2010, ASML shipped the first prototype EUV tool to an undisclosed Asian customer. Semiconductor production today is divided into the EUV “haves” like Taiwan Semiconductor Manufacturing Co. (TSMC), Samsung and Intel, which make advanced chips for customers like Apple, MediaTek and Qualcomm. The EUV “have not” chip makers years ago threw in the towel at leading nodes, jettisoning the associated multi-billion dollar capital expenditures and focusing on improved profits from legacy production lines and products that benefit little or none from process shrinks.
Europe’s new €1.6bn chip plant needs only 10 workers on factory floor
A 60,000 square meter facility built specializing in power semiconductors seeks ease bottlenecks for major automotive clients. The increase in automation solutions has made localized European production of semiconductors possible. By reducing comparable personnel needed to run the facility from 150 to 10 makes the factory cost competitive with factories in Asia.
Fabs Drive Deeper Into Machine Learning
For the past couple decades, semiconductor manufacturers have relied on computer vision, which is one of the earliest applications of machine learning in semiconductor manufacturing. Referred to as Automated Optical Inspection (AOI), these systems use signal processing algorithms to identify macro and micro physical deformations.
Defect detection provides a feedback loop for fab processing steps. Wafer test results produce bin maps (good or bad die), which also can be analyzed as images. Their data granularity is significantly larger than the pixelated data from an optical inspection tool. Yet test results from wafer maps can match the splatters generated during lithography and scratches produced from handling that AOI systems can miss. Thus, wafer test maps give useful feedback to the fab.
The Big Semiconductor Water Problem
The $150 Million Machine Keeping Moore’s Law Alive
ASML’s next-generation extreme ultraviolet lithography machines achieve previously unattainable levels of precision, which means chips can keep shrinking for years to come.
ASML introduced the first extreme ultraviolet (EUV) lithography machines for mass production in 2017, after decades spent mastering the technique. The machines perform a crucial role in the chipmaking ecosystem, and they have been used in the manufacture of the latest, most advanced chips, including those in new iPhones as well as computers used for artificial intelligence. The company’s next EUV system, a part of which is being built in Wilton, Connecticut, will use a new trick to minimize the wavelength of light it uses—shrinking the size of features on the resulting chips and boosting their performance—more than ever before.
Semiconductor Manufacturing: Making Impossibly Small Features
Deposition, litho, and etch are interdependent processes tightly linked together. Technology continues to press forward to meet the incredible opportunity of 5G, cloud, and IoT, and these processes enable our customers to go further by reducing feature sizes and improving pattern fidelity to get a better feature. The litho process can produce lines that have some edge roughness, which can have a negative impact on the variability of the devices you make because it affects what are called “critical dimensions,” such as the transistor gate length. We can also take litho printed lines and improve the sidewall smoothness by typically 30% or more.
Aiming for the Top in Industrial AI, SK’s First AI Company Gauss Labs
Gauss Labs has been developing AI solutions aimed at maximizing production efficiency using the massive amount of data generated at SK hynix’s production sites. SK hynix wishes to make the overall semiconductor production process more intelligent and optimized across all procedures including process management, yield prediction, equipment repair and maintenance, materials measurement, and defect testing and prevention.
MacroFab: Driving The Cloud-Based Transformation Of Electronics Manufacturing
The company brings cloud-based, manufacturing-as-a-service (MaaS) solutions to the electronics industry. On its platform, companies can upload component designs, obtain quotes, place orders and follow the progress towards delivery. Companies can price and order a wide range of parts and products, from printed circuit boards (PCB) to fully assembled and packaged electronics products.
AI In Inspection, Metrology, And Test
“The human eye can see things that no amount of machine learning can,” said Subodh Kulkarni, CEO of CyberOptics. “That’s where some of the sophistication is starting to happen now. Our current systems use a primitive kind of AI technology. Once you look at the image, you can see a problem. And our AI machine doesn’t see that. But then you go to the deep learning kind of algorithms, where you have very serious Ph.D.-level people programming one algorithm for a week, and they can detect all those things. But it takes them a week to program those things, which today is not practical.”
That’s beginning to change. “We’re seeing faster deep-learning algorithms that can be more easily programmed,” Kulkarni said. “But the defects also are getting harder to catch by a machine, so there is still a gap. The biggest bang for the buck is not going to come from improving cameras or projectors or any of the equipment that we use to generate optical images. It’s going to be interpreting optical images.”
Edge-Inference Architectures Proliferate
What makes one AI system better than another depends on a lot of different factors, including some that aren’t entirely clear.
The new offerings exhibit a wide range of structure, technology, and optimization goals. All must be gentle on power, but some target wired devices while others target battery-powered devices, giving different power/performance targets. While no single architecture is expected to solve every problem, the industry is in a phase of proliferation, not consolidation. It will be a while before the dust settles on the preferred architectures.
The misplaced optimism is twofold: first there is the fact that eight years later Intel has again appointed a new CEO (Pat Gelsinger), not to replace the one I was writing about (Brian Krzanich), but rather his successor (Bob Swan). Clearly the opportunity was not seized. What is more concerning is that the question is no longer about seizing an opportunity but about survival, and it is the United States that has the most to lose.
Early And Fine Virtual Binning
ProteanTecs enables manufacturers to bin chips virtually, in a straightforward and inexpensive way based on Deep Data. By using a combination of tiny on-chip test circuits called “Agents” and sophisticated AI software, chip makers can find relationships between any chip’s internal behavior and the parameters measured during the standard characterization process. Those relationships can be used to measure similar chips’ internal characteristics at wafer sort to precisely predict how chips would perform during Final Test, even before the wafer is scribed.
Early And Fine Virtual Binning
ProteanTecs, which provides an AI platform to monitor chip reliability, today closed a $45 million funding round. The company says the fresh capital will bolster its go-to-market strategy and operations as it seeks to scale worldwide.
Chip design and manufacturing is a high-risk, high-reward pursuit. Mistakes made during the earliest phases are often enormously costly — chip fabrication plants cost billions to build. And the most sophisticated hardware can take years to hammer out, with intense speculation about how to optimize the next generation for workloads that might come into vogue.