Boosting Defect Inspection with Artificial Intelligence: Evolution of SCREEN's Optical Inspection Platforms

Semiconductor Review | Thursday, February 22, 2024

SCREEN Semiconductor Solutions Co., Ltd. headquartered in Kyoto, is a global leading equipment manufacturer for the semiconductor industry and actively operates in wet etch and clean, lithography materials coating and development, wafer annealing, and metrology.

On top, SCREEN has a portfolio of wafer inspection tools targeting silicon (Si), gallium nitride (GaN), silicon carbide (SiC), sapphire (Al2O3), lithium tantalate (LiTaO3), and lithium niobate (LiNbO4) substrates from 76mm to 300mm in diameter.

Stay ahead of the industry with exclusive feature stories on the top companies, expert insights and the latest news delivered straight to your inbox. Subscribe today.

SCREEN’s inspection tools are characterized by high throughput, small footprint, and low cost-of-ownership (COO) and will deliver new inspection solutions powered by artificial intelligence (AI). 

We met Alessandro Rossi, product manager, for a deep conversation on SCREEN’s inspection technology and solutions.

Alessandro, can you tell us more about SCREEN’s wafer inspection business?

More than 15 years ago, our process group started to identify the main yield-killer defects in wafer high-volume manufacturing (HVM) settings. We worked closely with our customers to identify the critical process steps and the associated defects responsible for yield drop.

The most common defects are generated by an abnormal dispense, due - for example - to air bubbles in the coater photoresist or development line, which brings about non-uniform coverage and de-wetting. Other types of defects, both random and repeated, can come from the exposure tool itself.

When we started our evaluation, we were dealing with a technology node of 0.35 um, and visual inspections were carried out on a limited sample of wafers by using manual or semi-automatic tools.

Nowadays we work in more advanced markets, often requiring the inspection of 100% of production wafers and -sometimes- dies. In this scenario, a fully automated inspection tool is a must-have.

Which are the SCREEN’s wafer inspection platforms? Which customer needs SCREEN intends to address?

We launched our first automated inspection system, ZI-2000, in 2010. This tool has been demonstrated to be suitable for many patterned and unpatterned wafer types, including Si, SiC, GaN, Al2O3, LiTaO3, and LiNbO4. The software allows an easy recipe setting, automated defect classification (ADC), and defect viewer analysis (DVA). Due to small dimensions, high throughput, and extreme user-friendliness, the product was well received by main high-volume manufacturers operating in semiconductor, optoelectronics, and power electronics.

Figure 1. ZI-2000 is an automated wafer inspection system from SCREEN Semiconductor Solution which handles different substrate types from 76mm to 200mm in diameter.

More specifically about power electronics, SiC is becoming the substrate of choice for power devices in the automotive and energy storage application space. SiC substrates and SiC epitaxial growth are still associated with many device-killer defects, the most popular being cracks, triangular, and “downfall” defects.   All these defects are potential device killers and must be detected at an early production stage. ZI-2000 auto defect classification (ADC) function has been proven to be effective on killer and non-killer defects classification, with consistent cost and time savings for the end-users.

Figure 2. An example of a defect library obtained with ZI-2000 on SiC epi wafers

More generally, the IoT market is rapidly growing and presents specific requirements. To meet this need, SCREEN has recently launched the ZI-3600 wafer pattern inspection system to support the stable production of power chips, MEMS, and CMOS image sensors.

Figure 3. ZI-3600 is the latest automated visual inspection (AVI) system from SCREEN Semiconductor Solution which handles different substrate types from 100mm to 300 mm in diameter.

Why should your customers adopt the ZI-3600 platform? Tell us a few convincing reasons!

There are at least 6 excellent reasons to adopt our tools!

1. Inspection boosted by artificial intelligence. All inspection results can be classified in real time.  Machine-learning-based algorithms are used for automated defect classification, based on pre-set grading parameters. We constantly push the limits of our technology to provide our customers with powerful, accurate, and reliable inspection tools. We recently released our AI-boosted false defect filtering system under a new brand name, SCRAIS. Combining our proprietary image processing technology with AI, we aim to reduce false calls by over 80% with an oversight rate below 0.1% on the customer side. Additionally, we reduce data labelling processes for AI modelling and optimise the setting procedures.

2. Easy inspection recipe management. The inspection recipe can be tuned “on the fly” during the defect acquisition itself. In the case of patterned wafers, the measurement of line and space critical dimension (L/S CD) and overlay can be carried out just after the inspection itself, avoiding a separate tool.

3. Industry-leading high resolution. ZI-3600 is equipped with three objective lenses of different resolutions. The lenses are automatically switched and allow the detection of defects 1μm or less in size.

4. Flexible inspection methods. The tool is equipped with long-life LEDs. Besides unpatterned wafers, three different types of inspection methods—die-to-die, cell-to-cell, and die-to-reference—can be combined within a single recipe.

5. Flexible configuration in a space-saving design. A variety of configurations, including open cassette only, open cassette and FOUP combined, FOUP only, SMIF, and AGV is possible on a compact footprint.

6. Doubling throughput of the previous model. Equipped with a proprietary inspection head and an improved high-speed image processing engine, the ZI-3600 doubles the practical processing capacity of previous models. It also features a line sensor to enable the inspection of the entire wafer surface. Wafers can be inspected at a constant speed regardless of chip size or quantity, greatly contributing to stable production. ZI-3600 is specifically recommended for inspection in a mass production environment, where a 100% inspection sampling rate is required, as – for example - in the automotive sector. 

SUMMARY

SCREEN offers automated unpatterned and patterned wafer inspection tools for silicon (Si), gallium nitride (GaN), silicon carbide (SiC), sapphire (Al2O3), lithium tantalate (LiTaO3), and lithium niobate (LiNbO4) ranging from 76mm to 300 mm in diameter size. Characterized by high throughput, small footprint, low cost-of-ownership, and boosted by AI-supported algorithms, SCREEN’s automated system provides a sensitive and reliable inspection to support customers operating in CMOS imager sensors, automotive, MEMS, photonics, and LED industries.

Feel free to address your product inquiries to Alessandro Rossi, SCREEN SPE Italy Srl. (a.rossi@screen-spe.eu)

 
 
 
 

More in News

The pursuit of absolute purity in semiconductor manufacturing presents significant challenges, as even a single microscopic particle can jeopardize the integrity of an entire production batch. As integrated circuits continue to shrink to the atomic level, the performance specifications governing the packaging and transportation of aggressive chemicals used in etching and cleaning processes have become increasingly stringent. Advanced fluororesins have emerged as fundamental components of this infrastructure, offering exceptional thermal stability and chemical inertness, attributes unmatched by any alternative materials. These application-specific polymers establish a protective barrier against contamination, thereby facilitating global fabrication facilities in their efforts to achieve higher yield objectives. When integrated into the production line, these high-performance materials provide manufacturers with a competitive advantage characterized by enhanced reliability and long-term structural integrity. How Does Chemical Inertness Drive Process Stability? The main advantage of fluororesins in semiconductor applications is their resistance to almost all corrosive materials used in wafer processing. Strong acids and organic solvents can corrode regular piping and storage systems, thus releasing metallic ions or particulates into the process stream. Under such conditions, fluororesins do not react, thereby preserving the ultra-high purity required for the processing of advanced logic and memory chips. This stability directly translates into reduced downtime for equipment maintenance and a corresponding reduction in the risk of unforeseen system breakdowns. Diversified Technologies, Inc. implements comparable high-purity and contamination-control measures in semiconductor systems to ensure consistent operational reliability. Due to their low surface energy, these materials reduce residue buildup and facilitate easier drainage, thus providing the shortest possible interruption for tools returning to production between flushes. As production facilities continue to improve their resource consumption, the long lifetimes of fluororesin parts open a straightforward approach to sustainable operational efficiency and lower total cost of ownership throughout the equipment lifecycle. DEMCO provides industrial-grade components and materials designed to maintain process integrity and minimize downtime in advanced semiconductor manufacturing. Why Is Thermal Resilience Crucial For Scaling? As the industry moves toward more complex three-dimensional architectures, the thermal loads on components being manufactured are increasing. Fluororesins would retain their mechanical properties over a wide temperature range, which is necessary in setups with very rapid thermal cycling or chemicals washed at high temperatures. Their thermal resilience ensures that seals, gaskets, and linings will not deform or lose their sealing properties in response to heat, thereby maintaining vacuum integrity in the processing chamber. It is precisely this ability to withstand changes without structural damage or outgassing that allows engineers to push the current limits of fabrication. In addressing material selection from a strategic viewpoint, business leaders must acknowledge that the physical limitations of the manufacturing environment are often equilibrated by the materials that constitute the containment of the process. The investment in advanced fluororesin solutions provides organizations' technical teams with opportunities to explore the cutting edge of microelectronics. This convergence of material science and industrial strategy will ensure that the infrastructure of the digital economy remains robust and poised to meet the next generation of consumer and enterprise demand for processing power. ...Read more
The proliferation of the Internet of Things (IoT) and advanced embedded systems is fundamentally reshaping the digital landscape. From smart cities and connected vehicles to industrial automation and intelligent medical devices, these innovations are driving an unprecedented demand for localized intelligence and real-time processing. At the heart of this revolution lie semiconductors, the unsung heroes powering the "edge" – where data is generated and acted upon, rather than solely relying on distant cloud infrastructure. The Semiconductor Imperative: Core Requirements for the Edge To meet the demanding requirements of next-generation IoT and embedded systems, semiconductors are evolving rapidly, with a strong focus on key performance attributes. Foremost is low power consumption, as many IoT devices operate on batteries in environments where frequent recharging or replacement is not feasible. This necessitates ultra-low-power architectures, efficient RF and physical layer designs, and optimized deep-sleep modes. Innovations such as voltage and frequency scaling (DVFS) and smart memory integration play a critical role in extending battery life. The shift toward processing complex AI and machine learning tasks at the edge has also increased the need for high-performance, energy-efficient computing. To this end, specialized processors—including AI accelerators and Neural Processing Units (NPUs)—are being integrated to enable real-time inference with minimal power usage. Miniaturization and functional integration are also vital, as IoT devices are often constrained by size and cost. Advances in semiconductor design now enable the consolidation of processing, memory, connectivity, and security functions into a single System-on-Chip (SoC), thereby improving efficiency across the board. Equally important is robust security. As IoT devices become more widespread, they are increasingly vulnerable to cyber threats. Semiconductor designs are therefore incorporating hardware-level security features such as secure boot, hardware root of trust, encryption for secure communication, secure key management, and tamper detection mechanisms. Innovations Driving the Future of Edge Semiconductors The semiconductor industry is experiencing a surge of innovation aimed at meeting the demands of next-generation computing. One significant development is the emergence of edge AI chips—dedicated processors designed to perform inference directly on devices. These chips are optimized for applications such as surveillance, industrial robotics, and smart home systems, with a focus on power efficiency, low latency, and compact form factors. Likewise, chiplet architectures are gaining prominence, replacing traditional monolithic designs with modular components that enhance scalability, enable efficient power management, and support heterogeneous integration within a single package. Advancements in process technologies are also driving progress, with leading-edge fabrication techniques like FinFET and Gate-All-Around (GAA) transistors improving energy efficiency by reducing power leakage and lowering operating voltages. Meanwhile, the adoption of the open-source RISC-V instruction set architecture is expanding across embedded systems. Its flexibility and cost-effectiveness enable greater customization, thereby accelerating innovation in chip design. Another transformative area is neuromorphic computing, which mimics the architecture of the human brain to deliver ultra-low-power, high-performance AI processing. This approach holds particular promise for real-time analytics and adaptive, self-learning embedded systems. The symbiotic relationship between semiconductors and edge computing will continue to drive innovation. As the IoT expands exponentially, the demand for more intelligent, secure, and energy-efficient edge devices will only grow. The future of embedded systems and IoT will be defined by semiconductor advancements that enable faster processing, robust security, longer battery life, and seamless connectivity, truly empowering the edge to become a strong, distributed intelligence network. ...Read more
Executives acquiring quartzware fabrication systems face a purchasing decision that reaches beyond component replacement. In semiconductor and adjacent manufacturing environments, quartzware sits close to process performance, equipment uptime and yield protection. A tank, carrier, tube, chamber or heat exchanger that appears minor on a procurement list can become a constraint when purity, tolerance control or lead time fails to match production needs. The right fabrication partner must therefore be assessed not only by what it can build, but by how consistently it can support continuity when demand shifts, equipment ages or legacy parts become difficult to source. Current pressure on manufacturing teams has made this decision more exacting. Many facilities are trying to extend asset life while maintaining process discipline, controlling cost and avoiding downtime. New builds remain necessary, but replacement alone is not always the rational answer. When a damaged quartz component can be restored to specification, repair capability becomes a strategic purchasing advantage. It helps teams preserve scarce capital, shorten recovery windows and reduce dependence on fresh fabrication when production schedules are exposed to raw material and capacity delays. Precision is still the baseline. Quartzware used in semiconductor equipment must be fabricated from appropriate high-purity material and held to tight dimensional requirements because contamination and variation can travel quickly into yield loss. Buyers should look for evidence that the supplier treats quality control as a release discipline rather than a final formality. That means documented inspection, tolerance verification, equipment upkeep and a workforce trained to repeat complex work without relying only on individual memory. A supplier’s ability to keep machinery calibrated, maintain fixtures and invest in the people doing the work often determines whether quality holds steady beyond one successful order. Flexibility is equally important. Many manufacturers inherit tools, assemblies and parts without current drawings, especially when equipment has been modified, transferred or supported by older OEM documentation. A capable quartzware fabrication system should be able to reverse engineer unfamiliar parts, translate samples into workable drawings and support both one-off and recurring needs. This is where the distinction between a parts vendor and a true fabrication resource becomes clear. The stronger partner can handle unusual geometries, repairs, production pieces and urgent requests while preserving inspection discipline and delivery reliability. Delivery performance should be treated as a technical requirement, not a customer service promise. In fabricationdependent manufacturing, lateness does not merely inconvenience purchasing. It can idle equipment, delay qualification or force teams into lower-quality substitutes. Strong suppliers build reliability through planned maintenance, trained labor, transparent quoting and realistic commitments. The best choice is the partner that can combine purity, repair judgment, engineering adaptability and delivery consistency into one dependable source. Desert Glass Works stands out for buyers needing this balance. Its relevant work spans quartzware repair, custom machining and precision fabrication for tanks, carriers, process chambers, diffusion tubes, bell jars, heat exchangers and related semiconductor components. The company’s model is well matched to buyers managing new requirements and legacy equipment, since it can repair or restore parts, reverse engineer components without drawings and fabricate custom quartzware from high-purity stock. Its emphasis on full inspection, maintained equipment, apprenticeship-based skill development and on-time delivery makes it a strong recommendation for organizations that need quartzware support tied directly to yield, uptime and disciplined execution. ...Read more
Longer validation cycles are forcing semiconductor factories to examine a cost that rarely appears cleanly on a procurement sheet: the time lost among test readiness, tool availability, engineering bandwidth and material movement. Advanced packaging, AI processors, automotive electronics and mixed-signal devices have widened the range of qualification work inside the same facility. A test cell may be technically capable, yet throughput still suffers when device programs shift faster than handlers and factory logistics can adjust. That pressure often shows up as idle equipment rather than an obvious planning failure. The buying question is no longer confined to tester performance. Executives responsible for semiconductor test and robotics systems need to understand how a platform behaves when product mix changes and validation data expands while factory movement becomes a constraint on output. Hardware flexibility matters because qualification programs cannot wait for major reconfiguration every time a device family changes. The test environment should support wafer sort, final test, system-level validation and diagnostic review while allowing engineering teams to move between device requirements with limited disruption.  Diagnostic visibility carries equal weight. AI accelerators and dense system-on-chip designs produce large volumes of validation data under demanding electrical conditions. Engineers need earlier anomaly detection, clearer failure analysis, yield-behavior context  and a practical way to connect test results to process variation. A system that only records pass-fail results leaves too much interpretation for later review. Better infrastructure helps teams identify variation while the qualification window is still open. Factory movement has become part of the same purchase logic. Semiconductor facilities often lose time in the handoff between test stages, especially when materials or qualified components depend on manual transport. Robotics should reduce that drag without forcing the factory into a redesign. Collaborative robots must be practical for inspection and machine tending, while autonomous mobile robots should coordinate transport across existing floor layouts, adjust to changing routes, respond to obstacles and reduce waiting time around qualification flow. Integration is where many automation programs lose momentum. Test equipment, robotics systems, factory software and data tools frequently come from different vendors, making communication gaps a real production issue. Buyers should press for open integration and clear status data. Deployment paths also matter, especially when custom work stretches past the point of early value. Reliability remains nonnegotiable, since a robot or tester that adds downtime during a qualification push creates the problem it was purchased to solve. Teradyne fits this buying logic because it brings semiconductor test equipment and intelligent robotics into one portfolio relevant to validation and factory flow. Its test systems support wafer sort, final test, system-level validation and broader device qualification for memory, analog, mixed-signal and system-onchip applications. Universal Robots gives it collaborative robots suited to machine tending and inspection work. Mobile Industrial Robots supports autonomous transport inside manufacturing environments. Teradyne’s emphasis on analytics, adaptable test platforms, collaborative automation and mobile robotics gives semiconductor executives a practical route to connect validation accuracy with steadier factory coordination. For buyers trying to reduce qualification delays without separating test decisions from movement constraints, it deserves close evaluation. ...Read more