The disruptive force of generative AI is causing a paradigm shift across various sectors. To strengthen Taiwan's generative AI power and broaden generative AI applications, DIGITIMES organized the AI Wave: Taiwan Generative AI Applications Hackathon.Many teams comprising students from high schools and universities as well as people working in various industries joined the competition. On the day of the final contest, experts from Amazon Web Services (AWS) shared their insights into generative AI technology trends. A panel discussion with industry leaders, government representatives, scholars, and researchers focusing on generative AI talent incubation was held to shed light on generative AI development trends and job opportunities.The AI Wave: Taiwan Generative AI Applications Hackathon attracted 61 teams comprising students from 29 high schools and universities as well as people working in various industries. After one month of training, the teams came up with generative AI solutions demonstrating both creativity and usability.The competition was divided into a hackathon division and an idea exchange division. The teams in the hackathon division worked on problems related to their business specialty and developed solutions in the AWS environment using their own business data. Participants in the idea exchange division could form their own teams and use AWS PartyRock, a no-code service, to focus efforts on developing creative ideas for generative AI applications.AWS experts looked into generative AI technology trendsIn a series of talks on generative AI technology trends, AWS experts pointed out that the ability to create, produce, provide insight, and enable innovative experiences are the core values of generative AI. The technology can generate whole-new content and ideas while boosting business productivity and can enhance customer and employee experiences through never-before-seen human-machine interactions. For example, using AWS AI/ML services, Amazon Pharmacy built an LLM-based chatbot that helps customer service agents answer questions more effectively and a price estimate tool that optimizes the user experience. Generative AI and machine learning advances are not possible without the support of high-performance computing and storage infrastructure. In their talks, AWS experts introduced AWS solutions from the perspectives of storage and computation.For storage, the integration of Amazon FSx for Lustre with Amazon S3 data lakes supports fast random data access and high throughput, significantly boosting training efficiency and reducing costs. For computation, AWS provides cost-effective AWS Graviton processors as well as AWS Inferentia and AWS Trainium accelerators purpose-built for deep learning and inference. On top of that, AWS also offers new services and tools such as Amazon S3 Connector for PyTorch to bring performance and cost benefits to an even higher level.Generative AI solutions require the support of powerful data structures. AI models are commonly developed through ways such as in-context learning, model fine-tuning, and building from scratch. AWS provides a slew of database and storage services to enable high-efficiency data management and search, wherein data pipelines and data lakes play a vital role.Generative AI applications rely on high-efficiency data management and search to operate without a hitch. AWS techniques and tools take generative AI performance and scalability to the next level, helping companies rapidly deploy and run generative AI applications, raise data management efficiency, and reap greater AI benefits.Harnessing the power of AWS to build end-to-end data strategies, companies can strengthen their ability to analyze and gain insights into their business. According to AWS experts, Amazon OpenSearch Service can perform highly efficient searches and analyses with significantly optimized relevancy and precision.The vector engine of OpenSearch searches billions of vector embeddings in just milliseconds. It is suitable for applications including recommendation engines, image recognition, and fraud detection. AWS also has data integration services that allow users to effortlessly synchronize data from Amazon DynamoDB and Amazon S3 to OpenSearch. Data management is not only much easier but also more efficient and precise. With the powerful data analytic capability and generative AI technology enabled by AWS, corporations can build comprehensive data strategies and implement data-driven innovations, thereby enjoying the prosperity they bring.Leaders from the industry and academia communicate with the AI generation about international talent incubationIn the panel discussion on generative AI talent incubation, leaders from the industry and academia engaged in conversations with today's AI generation. The moderator, Dr. Ju-Chuan Wu, CEO of Cloud Innovation School, Feng Chia University, noted that it is almost impossible for humans to live without technology in the ICT era. Cloud Innovation School of Feng Chia University was founded to actively increase the breadth and depth of cloud applications as well as the effects and benefits of talent incubation through continuing industry-government-academia collaborations.With regards to human resource demand, Allen Chen, VP of Platform and New Product Development, 104 Corporation, commented that AI talent is in extreme shortage in Taiwan. Only one out of five AI job openings get filled with a suitable candidate. AI professionals enjoy promising career growth.Chen noted AI is already everywhere and people who use AI should get familiar with all types of devices to transcend old ways of thinking. In the times of AI, the smartphones that employees carry around are like an AI assistant that enables instant interaction and collaboration as well as real-time information look-up and response. As such, having AI tools play a part in employees' everyday work is instrumental to cultivating AI application thinking.According to Dr. Charles C.N. Wang, Dean of Industry-Academia Collaboration, Asia University, the university fosters AWS-certified professionals through peer learning. It also works with companies in central Taiwan to give real-world problems to students while encouraging cross-sector collaborations.Among the students that Asia University educates, 70% apply their knowledge and skills for work and 30% can develop new technologies. They are encouraged to start thinking about their future career path in their junior year.Furthermore, attaching importance to developing students' global competence, Asia University actively recruits international students and promotes bilingual education. Apart from professional skills, Dr. Wang urged students to also make themselves proficient in math, programming languages and English while taking the initiative to learn different cultures and broaden their international horizons. Yen-Ting Lin, a Ph.D. candidate at the Department of Computer Science and Information Engineering of National Taiwan University and the developer of the Taiwan-LLM project, voiced his opinions from a student's perspective. Colleges focus on building students' ability to develop robust AI systems while the industry needs their employees to have the ability to apply AI technologies to solve real-world problems, said Lin.Students can apply the AI skills they learn at school to various fields to increase work efficiency. He also noted that while learning and developing AI, students should also get a clear picture of the current trends and identify the fundamental problems so that they can use AI to improve existing processes.When developing AI systems, aside from technical capabilities, developers need to pay attention to building users' trust in AI. AI will only exert maximum benefits when it is used properly to help overcome inherent challenges that different fields are faced with.The AI Wave: Taiwan Generative AI Applications Hackathon drew 61 teams of students from 29 high schools and universities, alongside professionals from diverse industries. Following a month of training, teams showcased innovative and practical generative AI solutions

Passive infrared (PIR) sensors have been around for years and are widely used in various devices including smart lamps, automotive radar, and monitoring systems.They are instrumental and indispensable to the modern world. However, their lower sensitivity and smaller range also limit the capability of the smart devices that use PIR sensors.Millimeter wave (mmWave) radar sensors that are widely used in smart cars are now getting increasingly adopted in smart devices including monitoring and smart cockpit systems after manufacturers started to provide non-automotive mmWave radar chips a few years ago at more economical pricing. Millilab, with a robust R&D capability and technological strength, wowed visitors at CES 2023. It has engaged in close partnerships with multiple international monitoring system suppliers. Having raised NT$50 million in March 2024, Millilab is poised to capture the immense opportunities in the global smart vehicle market.According to Millilab's founder and CEO - Bibben Lin, mmWaves are shorter in wavelengths. Sensors based on mmWave radar technology provide higher resolution and accuracy so they are embraced by smart cars with high safety standards.Infineon used to dominate the mmWave radar chip market, making them expensive and hardly used in monitoring systems sold at lower prices compared to smart cars. When Texas Instruments (TI) introduced low-cost mmWave radar chips, it spurred a rapid increase in mmWave radar sensor adoptions in non-automotive applications. Today, almost all mid-range and high-end monitoring systems priced at US$200 or higher use mmWave radar sensors.Recognized by Japan-based Socionext, Millilab targets the global market with technological strengthEyeing mmWave radar sensors' high accuracy and high-resolution characteristics and the opportunities arising from decreasing costs, Lin with abundant experience in developing communication devices decided to found Millilab with his business partners in 2019. Millilab's mmWave radar solutions are being used in monitoring systems for infants and seniors, child presence detection (CPD) systems, and driver monitoring systems (DMS).When the company was first founded, the Millilab team focused on developing algorithms based on TI mmWave radar sensors for a variety of applications. Millilab's robust R&D capability soon caught the attention of Socionext, a Japan-based mmWave radar chip supplier.Each boasting unique technologies in their own field, the two companies decided to join forces and have together developed total solutions and reference designs for a diversity of application scenarios that are embraced by customers across different sectors. Millilab has become one of the most trusted partners of Socionext and now independently develops algorithms based on Socionext chips. It is also the company that many global brands turn to for mmWave radar solutions.Lin noted that during the company's early development, the competition was fierce in the automotive mmWave radar sensor market. European and American brands dominated 80% to 90% of the market and the rest was divided between two China-based vendors.In comparison, at that time almost every monitoring system was based on the PIR technology. This presented tremendous opportunities for Millilab. Not only did Millilab become Socionext's partner for its R&D strength but it also secured a deal with Europe's largest car seat brand Maxi-Cosi to jointly develop a crib (baby) monitoring system. The result impressed Maxi-Cosi and more importantly, gave Millilab global visibility.Millilab and partners expand into the smart cockpit market with authorities mandating CPD starting 2025With the assistance from TTA and flyingVest Ventures, Millilab attended CES in 2023 and 2024 as well as London Tech Week in the U.K. and VivaTech in France. The participation at these events helped Millilab reach customers from different corners of the world.For example, a Europe-based automatic door manufacturer was impressed by Millilab's algorithm and therefore plans to collaborate with the company. Multiple Asia-based corporations have also engaged in partnerships with Millilab to provide mmWave radar solutions for their products targeting overseas markets. These projects significantly fueled Millilab's revenue growth.Going forward, Millilab will undertake a three-way strategy, said Lin. First, it will continue to attend tech exhibitions and work with TTA and flyingVest Ventures to increase its global visibility. Second, to win more customers across different sectors, Millilab plans to implement the algorithm developed for Socionext chips on TI chips, ensuring the same level of performance and reliability. Last, the EU mandating CPD starting in 2025 is set to spur the demand for CPD systems so Millilab will work with partners to capture preemptive opportunities in this market segment.On average, about 40 children die each year in Europe after being left alone in a car. To address this problem, Euro NCAP will award safety rating points for CPD in new passenger vehicle designs from 2025.In response, manufacturers are actively developing smart cockpit solutions. This presents a great opportunity for Millilab, boasting strong algorithm R&D capability, to penetrate the market segment. It plans to introduce a standard smart cockpit solution in April 2024 for automakers' PoC testing, followed by customized designs according to their requirements. Millilab expects this to contribute to strong revenue growth in the coming years.(Editor's note: The original article was published in TTA Magazine Issue 14. Read more startup stories in TTA Magazines.)Millilab founder and CEO Bibben LinPhoto: MillilabSchematic diagram of Millilab's CPD systemPhoto: Millilab

Inevitably, the adoption of generative AI and LLM (Large Language Model) has fueled more intelligence and efficiency in our lives and works than in the past decade.Moving forward, the demand for running AI and LLM at the edge devices is growing as it provides lower latency, higher privacy, and more flexibility. It is set to redefine the level of intelligence of smart devices as well as broaden the landscape of mobile scenarios.Fibocom (Stock code: 300638), a global leading provider of IoT (Internet of Things) wireless solutions and wireless communication modules, launches a series of on-device AI solutions powered by Qualcomm QCS8550 and QCM6490 processors from Qualcomm Technologies, Inc. The solutions are designed to satisfy compute-intensive application scenarios such as robotics, automated vehicles, video collaborations, smart retailing, etc., accelerating industrial digitalization and intelligent transformation.Flagship on-device AI solution powered by Qualcomm QCS8550 processorUtilizing the powerful Qualcomm QCS8550 processor, Fibocom's flagship on-device AI solution is designed to deliver strong performance and unparalleled multimedia capabilities. Equipped with an octa-core CPU and an Adreno 740 GPU, the solution can support up to 4 concurrent displays, and 8K video encoding and decoding. It serves as a strong core for industries requiring high-definition video playing, fast data analytics, and lower latency like automated vehicles, robotics, remote medical surgery, computer vision systems, live streaming, video conference systems, and more.Premium on-device AI solution powered by Qualcomm QCM6490 processors, piloting the high-end AIoT marketThe solution developed using the Qualcomm QCM6490 processor, features an octa-core processor with high-speed HVX (Hexagon Vector Extension) technology, and a high-performance graphics engine to allow smooth 4K video playing and multi-channel camera inputs. In addition, the solution is capable of allowing a maximum of 5 ISPs (Image Signal Processing) and up to 5-8 camera streams simultaneously, helping customers to ease their concerns on multi-camera deployments as well as dual-screen display scenarios.The solution offers flexible wireless connections such as 5G, Wi-Fi, Bluetooth, and is equipped with a GNSS receiver for precise navigation both indoors and outdoors. In terms of software, the solution supports the mainstream operating systems: Android, Linux, and Windows. Leveraging the computing power of up to 13 TOPS, the solution efficiently helps customers handle data-intensive computation and processing, running various 1.3B/3B/7B open-source LLMs on the device, making it an ideal solution for smart retail, In-Vehicle Infotainment (IVI) and industrial inspection."We are excited to see our powerful Qualcomm processors, the QCS8550 and QCM6490, being utilized in Fibocom's innovative on-device AI solutions," stated Dev Singh, Vice President of Business Development and Head of building, enterprise & industrial automation at Qualcomm Technologies, Inc. "This collaboration is a testament to our commitment to advancing AI capabilities at the edge, enhancing performance and efficiency across a range of applications from industrial automation to smart retail.""It is paramount to master the productivity of AI and create value-added solutions from the edge side for our customers that are in urgent need of building their smart devices based on our solutions. We are thrilled to develop these solutions based on the advanced and powerful chipsets from Qualcomm Technologies, as it not only provides the fundamental architecture of edge intelligence, also enriches flexibility in network connections such as 5G/Wi-Fi/Bluetooth," said Ralph Zhao, VP of MC BU at Fibocom. "In collaboration with Qualcomm Technologies, Fibocom is dedicated to injecting new versatility to the future of intelligence, and accelerating the implementation of our collaboration in robotics, industrial automation, live streaming, and more."

Fibocom (Stock code: 300638), a global leading provider of IoT (Internet of Things) wireless solutions and wireless communication modules, launches 5G + Wi-Fi 7 solution for 5G FWA during Computex 2024, satisfying the demand for higher data throughput and extending Wi-Fi 7 performance in the home, SMB (Small and Medium-sized Business) and industrial scenarios.Boosting 5G + Wi-Fi 7 Capabilities of Peak Performance Up to BE7200The ultra-high bandwidth and low latency features make the FG370 an optimal wireless solution for 5G FWA scenarios like smart home, SMB, and industrial monitoring. The multiplier solution can be realized by coupling with the latest Wi-Fi 7 technology, unleashing the full potential of a lightning-fast wireless experience for end users.Compliant with the IEEE 802.11be (Wi-Fi 7) standard, the 5G + Wi-Fi 7 solution supports a 4K QAM (Quadrature Amplitude Modulation) modulation scheme as well as up to 160MHz channel bandwidth to expand the data transmission capacity and reduce latency efficiently. It is worth mentioning that the solution also supports dual-band 2.4GHz 4 x 4 and 5GHz 5 x5 to achieve MLO speed of up to 7.2Gbps.In terms of reliability and transmission efficiency, the integration with OFDMA, Multi-RU, and MU-MIMO technologies guarantees interference-free, signal-boosted performance. Driven by the unique 5T5R system empowered by the Wi-Fi 7 chipset, the end users will benefit from 5dB antenna gains and extended network coverage versus previous generations."As the world's first 5G module developed from MediaTek's T830, Fibocom FG370 has been recognized and deployed by global Tier 1 operators' 5G CPE commercialization since its first debut in October 2023. Evolved with the latest Wi-Fi 7 standard, Fibocom launched the tri-band BE19000 solution based on FG370 during MWC 2024," said Simon Tao, VP of MBB Product Management Dept., Head of MBB BU at Fibocom. "Up to date, Fibocom is glad to be the first module vendor launching the industry-first dual-band BE7200 solution based on FG370 compatible with Wi-Fi 7 chipset, significantly boosting the overall performance in terms of speed performance, cost, and frequency regulatory requirements. The Tri industry-leading innovations prove Fibocom's product capabilities on the MediaTek T830 platform and the dedication to helping customers win the market."Welcome to visit Fibocom stand #K1215a at Computex 2024 for more details about 5G + Wi-Fi 7 demonstrations.Fibocom extends market leadership in 5G FWA by launching 5G + Wi-Fi 7 solution at Computex 2024Photo: Company

Fibocom (Stock code: 300638), a global leading provider of IoT (Internet of Things) wireless solutions and wireless communication modules, today announces the expansion of its RedCap module portfolio by launching FG332 during Computex 2024 and introduces a cutting-edge 5G RedCap CPE solution integrated with the newly launched FG332 and the latest Wi-Fi 7/Wi-Fi 6 technologies.The industry-piloting layout of the 5G RedCap module and RedCap CPE solution is set to accelerate the 5G RedCap commercialization. The FG332 is a 3GPP R17-compliant module tailor-made for the lightweight of 5G IoT applications with optimized power consumption, enhanced network coverage, and R17 small data transmission.Developed from MediaTek's T300 chipset, the world's first 6nm Radio Frequency System-on-Chip (RFSOC) single die solution for 5G RedCap, the module supports 5G SA and a maximum bandwidth of 20MHz, enabling the peak transmitting performance up to 227Mbps downlink and 122Mbps uplink, and it is also backward compatible with 4G, ensuring the constant network connections. In hardware excellence, FG332 adopts the LGA form factor packaged at 29 x32mm and utilizes the cutting-edge hardware density design for a compact PCB area, allowing greater flexibility in product design.In terms of pin-compatibility, the module is pin-compatible with Fibocom LTE Cat 4 module series NL668, thus easing the customer's concern about hardware investment and time to market. Leveraging the optimized power consumption system, FG332 performs in a highly power-efficient mode and helps the UE (User Equipment) to extend the battery life. The module supports regional frequency bands in Europe, North America, and APAC, and the engineering sample of FG332-NA will be available in Q3 2024.FWA is gaining a larger market share to enable everyone's access to broadband networks anywhere, anytime. The FG332 is positioned for the mass deployment of CPE (Customer-Premises Equipment) and portable mobile hot spot devices with RedCap capabilities.Propelling the high-cost performance and low-power consumption of 5G RedCap, in conjunction with the latest Wi-Fi 7 and Wi-Fi 6 technology, Fibocom further offers a complete 5G RedCap CPE solution integrated with the latest Wi-Fi 7 and Wi-Fi 6 technologies, pivotal for consumer-level and enterprise-level users to enjoy the deadzone-free and interference-free wireless experience.The new generation 5G RedCap plus Wi-Fi 7 solution delivers higher data throughput of BE3600 and lower latency, compatible with IEEE 802.11a/b/g/n/ac/ax/be protocols, enabling the end devices to run at full speed of a maximum of 3571Mbps under the concurrent dual-band circumstance of 2.4GHz + 5GHz. The solution features Wi-Fi 7 benefits 4K-QAM for mass data loading thus enhancing the throughput, along with MLO, MRU, and preamble puncture to fully utilize the bandwidth, and software intelligence to improve the data transmission efficiency.For a more popularized 5G RedCap + Wi-Fi 6 selection, the solution utilizes a hardware accelerator for power optimization while maintaining an incredible speed of 3Gbps. The solution outperforms in both 2.4GHz and 5GHz frequency bands, the maximum 2402 Mbps can be achieved under the working mode of Wi-Fi 2X2 160MHz at 5GHz as well as 574Mbps under the working mode of Wi-Fi 2X2 40MHz at 2.4GHz. In terms of network coverage, uncompromised signaling and consistency will bring stable and seamless connectivity to end users."We are excited to unveil the FG332 RedCap module powered by MediaTek's T300 chipset at Computex 2024. At Fibocom, we are always dedicated to leading the way in 5G innovation, thanks to the cooperation with our ecosystem partner MediaTek, we launched the 5G RedCap CPE solution that integrates advanced Wi-Fi 7 and Wi-Fi 6 technologies, delivering exceptional data throughput and low latency," said Simon Tao, VP of MBB Product Management Dept., Head of MBB BU at Fibocom. "These innovations underscore our commitment to driving 5G RedCap monetization, ensuring our customers develop the FWA devices based on the latest technologies and capture the market opportunity ahead of time."Welcome to visit Fibocom stand #K1215a at Computex 2024 for more details about 5G RedCap demonstrations.Fibocom FG332

Detection not only safeguards the quality of products, but it can also discover flaws early in the production line through detection in key links – and further prevents the waste of manpower from continuous processing of defective products.The smart factory solution from ASUS AIoT which integrates AI, optics, IoT, and other technologies, has helped many manufacturer customers from different fields optimize the efficiencies of their production lines. Recently, this solution has been launched overseas for the first time to help a major motherboard manufacturer in China, Amertek Computer, Co, build a smart detection system to help the company reduce detection time and optimize the scheduling of human resources.ASUS established the ASUS IoT business group in 2019, and ASUS as a whole now retains nearly 1,000 engineers dedicated to the development of IoT and AIoT hardware and software. Under the framework of Industry 4.0, ASUS IoT develops B2B IoT and AIoT hardware and software integration services.Amertek was founded in 1994 and mainly manufactures computer mainboards. Its customers include international manufacturers of large servers and industrial computers. For its motherboard production lines, Amertek relied on manual visual inspection to detect flaws such as improper placement, missing components, skewness, and polarity reversal of components on the dual-inline package (DIP) line. This operation method produced has several pain points that are not easy to overcome.For starters, inspection personnel need to focus on tiny and complex components for long periods of time, which can easily cause visual fatigue and affect the accuracy and consistency of inspections. Manual visual inspection also relies on personal judgment, and different personnel may have different inspection standards, resulting in inconsistent inspection results.Then, when production quantity is increased, the efficiency of manual visual inspection is low and is unable to meet rapid production needs. Subsequently, when missed components enter the next production link and then problems are discovered, they need to be reworked or repaired, which causes increased cost and prolonged production cycles, affecting the production pace and product delivery times directly.Finally, the professional requirements for visual inspection works are relatively high and require long-term training and accumulation of experience to reach the inspection standard. In addition, due to the monotonous nature and high pressure of the work, the turnover rate of the staff is also greater compared to other workstations, and this causes the recruitment and training costs of enterprises to remain high.In order to soothe these pain points and more, Amertek decided to introduce a smart factory solution from ASUS AIoT. Since the production systems and processes of each manufacturing field are different, the ASUS IoT team provided a comprehensive service – from specification to build and installation.Numerous meetings were held during the early stages of system introduction to evaluate the specific requirements of Amertek and restrictions of the existing production processes. ASUS IoT engineers also visited Amertek's manufacturing site to understand the conditions of the production lines.Then, an AIoT solution was tailored according to the specific process requirements of Amertek, including customization of software and hardware tools, industrial camera configurations, optimization of image-processing algorithms, system integration and adjustments, and more. The ASUS IoT system accurately responded to the specific challenges of the production lines.Under the service of the ASUS IoT team, Amertek's current AI vision-inspection solution, AISDIS-100D, is equipped with industrial cameras that can scan both the front and back sides of motherboards at the same time. This solves the problem of manual visual inspection being able to inspect only one side of the mainboard, often resulting in missing the flaws on the other side.In addition, the AI algorithm developed by ASUS IoT can analyze the images collected by the industrial cameras and compare them with the images of the qualified and defective products in the database, vastly increasing the accuracy of inspection. The defect rate of the products on the production line was 3% during the early stages of introduction and had reduced to 1% after the introduction – significantly reducing material waste and human resources. The maintenance and rework costs of products also decreased.In terms of manpower scheduling, through the ASUS IoT AI vision-inspection solution, Amertek reduced the human resources required for inspection from the original two workers to one – and the transferees were moved to more satisfying jobs. By reducing necessary personnel by half, the inspection efficiency improved: the original manual operation took approximately 17 seconds per piece, and was cut to 15 seconds – not including the rest times required to alleviate eye fatigue. Also, the machine-learning model built into the ASUS IoT system can self-learn and adjust continually to quickly respond to the various production changes, maintaining the consistency of inspection quality for long periods of time.In terms of future plans, Amertek is currently cooperating with ASUS AIoT to develop an AI-powered review system that can be used for complex quality-control tasks such as precise flaw classification and detailed error analysis. Also, for specific problems such as mainboard deformation, Amertek plans to introduce professional board inspection systems to ensure that the products shipped from the factory can meet the highest physical specification standards.From the long-term goal and strategy perspective, Amertek plans to continue to expand and introduce the applications of ASUS IoT vision-inspection solution into its production lines – deepening its cooperation with both ASUS IoT and the wider ASUS group to expand its market competitiveness.For more product information, please refer to iot.asus.com.ASUS AISDIS 100D, equipped with industrial cameras capable of simultaneously scanning both sides of a motherboard and a proprietary AI algorithm, significantly improves inspection accuracy and reduces waste in materials and labor costs

Automated Optical Inspection (AOI) has always played a crucial role in quality control during the production process.In recent years, the proportion of high-mix, low-volume production modes has gradually increased, making traditional AOI technology insufficient to meet the current production line requirements. In response, industrial computer giant NEXCOM has adopted an AI-powered, vision-based smart factory solution from ASUS IoT.This integration has successfully applied AI-powered vision-inspection technology to the production of high-end industrial computers and servers. The system leverages AI to overcome the limitations of traditional AOI, significantly reducing modeling time and improving detection accuracy.ASUS established the ASUS IoT business group in 2019, and ASUS as a whole now retains nearly 1,000 engineers dedicated to the development of IoT and AIoT hardware and software. Under the framework of Industry 4.0, ASUS IoT develops B2B IoT and AIoT hardware and software integration services.NEXCOM, a prominent industrial manufacturer for more than 30 years, primarily produces industrial mainboards, high-end industrial computers, and AI servers. Its high-end industrial computers are called Box PCs, which are around the same size as makeup bags.Each Box PC contains up to 40 slots, each with 200 pins, resulting in a vast number of pins throughout the machine. In addition, with the development of AI servers, the number of CPUs and pins on the motherboards has also increased.NEXCOM previously relied mainly on manual visual inspection for quality control. However, it is difficult for the human eye to focus on tiny pins for long periods of time, and pin defects or foreign objects are challenging to detect from different angles.As the number and density of pins increase, the difficulty and uncertainty of manual inspection also increase. NEXCOM tried to introduce AOI equipment for automated inspection.However, since AOI equipment requires modeling for every pin defect, engineers need to write complex programs, which is both time-consuming and skill-intensive. Furthermore, the overkill rate – where good products are mistakenly identified as defective – of AOI equipment remains high.As a result, the workload of manual review did not drop. That's when NEXCOM called in ASUS AIoT to implement an AI-powered inspection solution to create a smart factory.As a professional industrial computer manufacturer, NEXCOM also established a subsidiary, NexAIOT, focusing on smart system integration. ASUS IoT has deep technical expertise in automated processes and equipment, especially in AOI and AI algorithms – with a substantial amount of implementation experience.The technical advantages of both parties highly complement one another. NexAIOT can efficiently integrate ASUS IoT solutions into NEXCOM's production environment, solving production line issues and creating new values.During the process of helping NEXCOM build its smart inspection system, the ASUS IoT team fully demonstrated its ability to provide customized services. Faced with the specific conditions of NEXCOM's production environment, including narrow spaces in production lines and large product sizes, ASUS IoT engineers team first conducted on-site evaluations to understand the specific needs and then optimized the system architecture accordingly.In terms of hardware design, ASUS IoT transformed the original floor-standing inspection equipment into a suspended type to fit NEXCOM's continuous production line layout. For the larger high-end industrial computer production lines, the inspection equipment was designed to be lightweight and movable, improving flexibility to meet the needs of different production lines and inspection points. In terms of software, ASUS IoT customized and adjusted the interface of the inspection software, including button positions, font sizes and warning colors to improve the operational convenience for users.ASUS IoT also helped NEXCOM accelerate the modeling efficiency of the inspection system through AI technology. Unlike traditional AOI that requires extensive time to write programs, the ASUS IoT AI inspection solution, AISDIS 100P NCC, requires only a small number of good product images with simple defect annotations to complete modeling within 30 minutes.This modeling method allows the inspection equipment to quickly go online in the early stages of new product introduction, perfectly matching the production pace. In addition, the deep learning algorithm of ASUS IoT AI inspection solutions can identify pin defects and foreign objects quickly and accurately, maintaining high levels of detection rate under different angles and lighting conditions.With the self-learning capability of the AI model, the system continually optimizes detection rules, keeping the overkill rate under 15%, significantly lower than the 70% of manual inspection. In terms of production efficiency, AI vision inspection also improved the inspection time significantly. In the past, manual inspection of an industrial computer took in excess of a minute, but the AI system reduces this to 30-40 seconds.Beyond solving existing problems, ASUS IoT AI inspection solutions also brought new values to NEXCOM. Through AI empowerment, NEXCOM has enhanced its flexible production capabilities to adapt to market changes and established a data-driven mechanism for precise manufacturing and continuous optimization. In terms of business, this has enabled NEXCOM to add ODM design services, successfully initiating its smart transformation.As for future plans, NEXCOM will continue to cooperate with ASUS IoT, expanding AI vision-inspection applications from motherboard and machine inspections to warehousing, logistics, and production management. This will create a comprehensive and multi-faceted smart manufacturing ecosystem, allowing the benefits of smart technology to become the driving force for the sustainable growth of the enterprise.For more product information, please refer to iot.asus.com.ASUS IoT's AI inspection solution, AISDIS 100P NCC, can rapidly and accurately identify pin defects and foreign objects, maintaining a high detection rate even under different angles and lighting conditions – reducing operating time by up to 50%Photo: ASUS

Ensuring the quality and efficiency of batteries is crucial for the performance and safety of electric vehicles (EVs). With this in mind, most new energy EV battery manufacturers are actively advancing towards smart technologies, integrating motion control, visual inspection, and other technologies into the initial, middle, and final stages of the manufacturing process.For instance, a leading automotive battery manufacturer has recognized that traditional manual sampling inspections are not only time-consuming and labor-intensive but also tend to identify defects late in the process. This results in the production of defective products and increased waste.In due course, the manufacturer entrusted a specialized System Integrator (SI) in visual inspection and AI analysis to establish a tailored quality control mechanism. This involves real-time detection and adjustment of "Key Process Input/Output Variables" (KPIV/KPOV) to prevent failure modes from affecting subsequent processes.However, given that the SI specializes in visual inspection software development rather than the installation and maintenance of computers and peripheral equipment, they sought a supplier capable of providing a one-stop solution that includes industrial computers, expansion cards, and GPU cards. This supplier must offer comprehensive after-sales service to address multi-source supply challenges, such as incompatibility and troubleshooting issues. After a thorough evaluation, the SI decided to adopt ADLINK Technology's complete solution, which includes an industrial PC, image capture card, and GPU card.Establishing a Digital Inspection Infrastructure with IPC and Image Capture CardsHsin-Chen Lin, Product Manager at ADLINK Technology's Edge Computing Platform Product Center, put an emphasis on the battery manufacturing process, which is divided into three stages: initial, middle, and final. The initial stage involves the "electrode production" process, which includes steps such as raw material mixing, coating, electrode pressing, and die-cutting.The middle stage is the "battery cell assembly," which involves placing the battery case through winding or stacking methods, followed by sealing through hot pressing, ultrasonic welding, and laser welding. The final stage is "activation testing," where the battery is activated by injecting electrolytes, followed by charge and discharge testing, and capacity grading and sorting marking the final step.Electric Vehicle Battery Manufacturing ProcessHe elaborated further that after the three stages, the battery proceeds to the modularization phase, which includes component gluing, side plate welding, harness installation, and final testing. These steps are pivotal for assembling the final automotive battery module.Using ADLINK Technology's IPC and professional image capture card visual inspection solutions for digital inspection enables real-time monitoring during these production stages. This ensures adherence to rigorous manufacturing standards at every step.Given the complexity of the over ten detailed processes across these stages, numerous critical monitoring and inspection requirements are presented. For example, in the early stage's coating process, AI technology with Camera Link cards and line scan cameras must accurately detect coating defects and width discrepancies, preventing conductivity issues and material exposure. It ensures the battery's optimal charge and discharge performance.Similarly, in the die-cutting process, it is essential to check the electrode tab-cutting precision. Lastly, the middle stage also presents strict challenges, ensuring precise alignment between winding and stacking and mitigating safety risks of low capacity or lithium plating. During ultrasonic and laser welding processes, it is necessary to meticulously inspect the electrode tab welds for scratches, fractures, and wrinkles. AI technology plays a key role here, effectively detecting defects that traditional vision systems may overlook.In the modularization stage, component gluing and shell welding processes also rely on advanced imaging technology and AI machine learning to inspect the quality which is crucial for preventing component misalignment and electrolyte leakage, thus avoiding potential battery short circuit issues.Understanding the specific inspection needs of SIs, ADLINK Technology delivers a complete solution based on the IMB-M47 motherboard as the core of the Industrial Computer (IPC), combined with a dedicated chassis and Power Supply Unit (PSU), paired with the PCIe-GIE74P image capture card and GPU card. This comprehensive solution showcases ADLINK Technology's thoroughness and expertise compared to most industrial control companies only offering motherboards or Box PCs.This one-stop solution simplifies system configuration and speeds up the deployment process. It allows SIs to quickly and effectively integrate visual inspection software into battery production equipment, significantly enhancing production efficiency and product quality control.Leading Electric Vehicle Battery Manufacturing into a New Era of InnovationADLINK Technology provides comprehensive support services, allowing SIs to confidently resolve any issues with industrial computers, expansion cards, and other critical components. This enables SIs to focus on software development without worrying about hardware-related problems, thereby helping automotive battery manufacturers reduce inspection time, lower scrap rates, and detect subtle defects. This results in lower production costs while enhancing battery production efficiency and quality.ADLINK solutions offering for EV battery manufacturingHsin-Chen Lin emphasized ADLINK Technology's latest solution, the IMB-M47 motherboard, highlighting its two significant technological advancements. He noted its superiority, achieved through the utilization of 12th/13th generation Intel Core processors combined with the latest memory and I/O expansion interfaces like DDR5 and PCIe Gen5, offering significantly better processing performance than the commonly used 8th/9th generation processors.It is complemented by the IMB-M47 motherboard's ability to support at least two PCIe-GIE74P image capture cards to handle multiple complex inspection tasks. With the increasing demand for visual inspection and the large data volume generated by high-resolution line scan images, the high-performance IMB-M47 motherboard paired with PCIe-GIE74P image capture cards ensures precise and fast inspections, making it an ideal combination.Comprehensive Expansion Card Solutions Accelerate Intelligent Battery ManufacturingADLINK Technology's greatest advantage in the realm of "intelligent manufacturing for new energy EV battery equipment" stems from its roots in expansion cards. Whether it's image capture cards, motion control cards, DIO cards, or GPU cards, ADLINK Technology boasts the most comprehensive and high-quality product lineup.Currently, the primary application involves integrating IPCs and image capture cards for visual inspection. However, across the entire automotive battery manufacturing process, spanning from early to late stages and including equipment such as mixers, coaters, cold press machines, die-cutters, winding/staking machines, and more, ADLINK Technology's innovative solutions play a vital role.Through ongoing R&D investments, ADLINK Technology consistently incorporates the latest technology standards and chipsets into its industrial computer systems, significantly boosting processing speed, data transfer efficiency, and reliability. Furthermore, ADLINK Technology's internal expansion card solutions effectively tackle compatibility issues faced by system integrators, further enhancing overall production efficiency.These innovative technological solutions reinforce ADLINK Technology's leadership in the automotive battery manufacturing sector and establish it as the preferred partner for automotive battery manufacturers, production equipment suppliers, and system integrators.Learn more about ADLINK Industrial PC or contact ADLINK for more information about products and solutions.

The global healthcare sector was amazed when a little girl with B-cell acute lymphoblastic leukemia was cured after being treated with chimeric antigen receptor (CAR) T-cell therapies (CAR-T) in 2012.Dr. Kuo-Hsiang Chuang, who just started his job at the Graduate Institute of Pharmacognosy, Taipei Medical University (TMU) at that time, thought CAR-T would help doctors and patients successfully fight cancer. However, CAR-T, mostly based on retrovirus gene delivery, leaves a lot of room for improvement. Dr. Chuang therefore developed the unique "Armed-T" technology and founded CytoArm in 2020, which now offers five T cell products against different cancers based on the Armed-T cell platform. CAR-T delivers effective results but has room for improvementAccording to a WHO report released on February 1, 2024, there were close to 20 million new cases of cancer in the year 2022 alongside 9.7 million deaths from cancer. It is not an exaggeration to call cancer the biggest threat to human health. This is also why the world including the healthcare sector is paying attention to CAR-T following the successful treatment of the little girl with leukemia in 2012.According to Dr. Chuang, CAR-T makes use of genetic engineering technologies to change the patient's immune cells called T cells, and enable them to effectively attack the cancer. T cells are collected from the patient and re-engineered in the laboratory to produce proteins on their surface called Chimeric Antigen Receptors (CAR) using retrovirus gene delivery.The genetically reengineered CAR-T cells can precisely identify tumor-specific antigens. When deployed back in the patient's body, CAR-T cells can specifically bind to cancer cells and destroy them.FDA approved the first CAR-T therapy in 2017, which has been proven to deliver promising results for blood cancers and impressive survival outcomes. However, Dr. Chuang pointed out the risks of CAR-T. First of all, it uses retrovirus vectors for gene delivery into T cells, which might elicit an unexpected immune response or cause T cell mutation.Furthermore, CAR-T costs will remain high as the processes of collecting, reengineering, and readministering T cells require a highly specialized team of professionals and expensive lab equipment. The gene (or virus) reengineering and cell culturing processes are also very challenging. Last but not least, CAR-T patents are owned by a small group of big pharma players, barring the entry of newcomers with formidable licensing fees or potential infringement lawsuits.Armed-T fights cancer even more effectivelyCytoArm's Armed-T technology addresses all of the above issues. According to Dr. Chuang, with Armed-T therapy, specific molecules are attached to the surface of the patient's T cell, much like putting armor on the T cell like Iron Man, described by Dr. Chuang.The armed T cells are then equipped to identify and destroy cancer cells. As opposed to CAR-T therapies that need 30 days for virus culture or reengineering, it only takes 10 days to produce high-purity cancer-fighting Armed-T cells and the therapeutic results are significantly more promising. To top it off, as no retrovirus gene delivery or genome editing technique is used, Armed-T cells are free of cancerization or mutation risks, making Armed-T a safer treatment choice for patients.In general, it is not easy for genetic reengineering techniques to achieve a high purity. CytoArm's Armed-T technology generates cancer-fighting cells at an impressive 90%+ purity. As to costs, without the need for retrovirus gene delivery, CytoArm can offer Armed-T technology at more affordable prices, allowing more patients to receive the treatment.Armed-T technology has another advantage, noted Dr. Chuang. The molecules that are attached to the surface of the T cell can be customized to allow the T cell to target and kill a specific type of cancer cell.CytoArm has developed five Armed-T products — CTA-01, CTA-02, CTA-03, CTA-04 and CTA-06. Experiments on lab mice with human B-cell leukemia indicate that CTA-01 can rapidly target and effectively destroy B-cell leukemia cancer cells with no obvious signs of recurrence. CTA-03 is for treating prostate cancer and CTA-04 is for treating metastatic breast carcinoma.CytoArm's first Armed-T product, CTA-02, is for treating incurable colorectal cancer associated with mutations in the epidermal growth factor receptor (EGFR) gene or its downstream effectors. In the first half of 2024, CytoArm is set to submit applications for CTA-02 clinical trials in Taiwan, Singapore, and the US.Poised to become a powerful weapon against cancer, CytoArm's Armed-T technology received worldwide attention following its introduction. Dr. Chuang named Taiwan's maturing startup ecosystem a key contributor to CytoArm's success.As a TMU-incubated startup, CytoArm developed a patent plan with TMU's guidance right when it was first founded. It also reached out to a U.S. IP firm for advice and confirmation that Armed-T does not infringe upon big pharma players' CAR-T patents, allowing the company to steer clear of obstacles that might hinder its research.The industry, government, and academia are stepping up efforts to support innovations and startups. Since CytoArm's establishment in 2020, capital investment, professional guidance, and market-driven strategies have all fueled the company's growth. Going forward, Dr. Chuang suggests that the government strengthen research grant and talent incubation policies and drive the commercialization of academic research to keep Taiwan's industry reinvigorated.CytoArm's Armed-T products. The first product, CTA-02, is for treating incurable colorectal cancer associated with mutations in the EGFR gene or its downstream effectors. Phase I clinical trials are scheduled to begin in 2024. It can be used to treat pancreatic cancer, lung cancer, TNBC and other types of cancer in the future.

AI technologies being used for image recognition is one of the disciplines seeing rapid developments today. AI image recognition is adopted in wide-ranging applications including city, factory, and traffic management systems.However, there are three major challenges to overcome to roll out AI image recognition implementations – limited service scalability, poor model adaptability, and time-consuming data collection. Linker Vision's all-in-one solution offers an auto-labeling tool to help system developers and application users reduce the need for human feedback. It also uses a data-driven AI training platform to increase model adaptability and scalability, thereby lowering cost and driving AI popularity. Established in 2021, Linker Vision focuses on full stack computer vision AI development, providing cloud-to-edge vertical integration solutions. Despite burgeoning AI and machine learning applications, businesses endeavoring to bring the applications into reality still face challenges.For example, they have difficulty scaling out their services. The models do not readily adapt to different use scenarios. Data annotation requires considerable time and effort. These problems hinder AI system implementation and rollout.To help businesses overcome these challenges, Linker Vision introduces the VisionAI platform. Based on Linker Vision's unique large vision model, VisionAI enables one-touch deployment of zero-shot AI, accelerates sampling and training for continuous learning, and provides auto-labeling for massive data annotation.VisionAI significantly speeds up the process of AI deployment and quickly boosts enterprises' core AI capabilities. Moreover, VisionAI reduces the human resources and costs required to have AI systems up and running.Linker Vision's solution has been widely used in autonomous driving, smart city, smart manufacturing, and a slew of other applications. The outstanding real-time image recognition performance provides strong support for enterprise digital transformation and automation upgrades. Comprehensive, convenient, and prompt services, Linker Vision leverages cloud computing for data storage and processing as well as model training of its AI solution. It also uses cloud platforms to quickly grow its service scale and application scope. Thanks to this, Linker Vision can create greater business value and capture enormous AI opportunities.