Thanks to the AI tsunami, demand for AI chips from all sectors is continuing to surge. The Chip-on-Wafer-on-Substrate (CoWoS) architecture that dominates existing 2.5D and 3D packaging technologies is however held back by the number of chips that can be produced on 12" wafers. Demand for AI chips from the explosion of AI applications will therefore continue to outstrip supply for at least the next few years.The industry is now turning to Panel-level Packaging (PLP) to overcome problems such as low yield and high cost in AI chip production. Increasing the usable wafer area will greatly increase the production output of AI chips. It is true however that many challenges still remain in the commercialization process for PLP, be it the production process or materials.This is why Manz, a PLP pioneer and leading supplier of RDL production equipment, proposed the new Chip-on-Panel-on-Substrate (CoPoS) concept to achieve a breakthrough in production. Manz drew on its many years of experience with PCB, IC substrate, panel, and packaging processes to come up with the "CoWoS panelization" solution. Under the CoPoS packaging architecture, chips are laid out on a rectangular substrate instead of a circular substrate to eliminate a bottleneck in AI chip production.Panel-level packaging is the key to overcoming the AI chip production bottleneckManz Asia General Manager Robert Lin stated that advanced packaging architectures generally have three components : the Chip Layer, Silicon Interposer, and IC Substrate. Continued reductions in IC pin size and line width make finding the right substrate difficult. RDL technology is therefore needed to redistribute pin locations and make lines wider so they can be successfully connected to the PCB. RDL is therefore crucial to improving chip packaging yields.What future challenges will CoWoS face? As new generations of AI chips appear, the number of dies in a single AI chip package will naturally increase as well. The overall footprint of the packaging will naturally increase. For CoW AI chips of mainstream brands, their footprint will grow by 50x54mm in 2023 all the way to 66x68mm in 2026, and 80x80mm in 2027. The number of AI chipsets produced per 12" wafer will drop from 14 to 11 then 4 in response. The ongoing decline in CoW production output will continue to inflate the production costs of AI chipsets as well.According to Lin, 700x700mm, 600x600mm or even the smallest 510x515mm panel-level packages all offer several times the usable area of 12" wafers. Chip layouts are also easier to develop for the square substrates. It is therefore not hard to see why IC companies all believe that PLP will be the future and this will create a tremendous opportunity for CoPoS.Leverage experience with RDL R&D and mass production to overcome the challenges in the PLP processUnder the panelization concept, the interposer in the packaging is produced as panels instead of wafers. The IC substrate is also changed from organic substrate to glass substrate. Intel had previously announced in 2023 that glass will be the next-generation substrate. The Through Glass Vias (TGV) process will be used to form the conductive layer that connects the substrates of the chip's upper and lower circuits. Glass was chosen because it is more resistant to warpage and signal noise than organic substrates, and offers superior thermal dissipation.Manz is a long-time developer of RDL production equipment and is now developing production equipment for conductive layers on glass substrate. A full set of wet process equipment for cleaning, developing, etching, stripping, and electro-plating has now been progressively developed. Manz is also using its extensive experience to shorten the learning curve for the industry and help them realize the mass production of glass substrates faster.Transitioning from a circular to square packaging process requires extensive investment in R&D on materials and equipment. Lin believes that since panels have a larger area than wafers and very different production processes, the CoPoS concept is no doubt one of the best solutions to the problem of production capacity; on the other hand, problems such as warpage and evenness must still be overcome to achieve high-definition conductor patterns and this will represent a challenge for yield.The above challenges highlight the advantages offered by Manz and new opportunities. Manz has been investing in panel-level packaging process equipment for quite some time. The Company has not only accumulated extensive experience with the delivery of mass production solutions but is also familiar with customer concerns regarding warpage and evenness. For this reason, Manz works closely with electroplating solution, glass and other material suppliers to help customers resolve related production issues as quickly as possible. Manz has already developed key production modules for Fan-Out Panel Level Packaging (FOPLP) and TGV that are now being validated by customers for pilot and mass production.Manz's next move will be to conduct further R&D on breakthroughs in RDL layouts in order to meet customers' exacting requirements on minimization of line width/spacing for AI chips. Manz is aiming for a reduction from 10µm to 5µm or even to 2µm and 1µm in its role as a supporter of advanced processes.Manz has continued to develop new breakthroughs that will provide customers with even better quality products and servicesPhoto: Manz

The annual OCP (Open Compute Project) Global Summit held every fall in San Jose, California, is a significant event for the global open computing community, bringing together industry leaders, researchers, and pioneers to discuss the latest developments in OCP projects and to share future challenges and innovation opportunities. The 2024 OCP Global Summit, taking place from October 15 to 17, centers around the theme "From Ideas to Impact," emphasizing OCP's commitment to accelerating responses to emerging trends and market demands while promoting openness, efficiency, sustainability, and scalability within the data center industry.At this year's summit, a competitive atmosphere filled the venue as exhibitors showcased their latest server solutions. As a leading global server chassis manufacturer, Chenbro made a significant impression by unveiling its latest server solutions based on the OCP DC-MHS (Data Center Modular Hardware System) architecture and AI server chassis products based on NVIDIA's MGX architecture. Chenbro's new products, co-developed with various system vendors and branded manufacturers, were prominently displayed across multiple booths at the event, underscoring the company's dedication to accelerating the development and time-to-market of next-generation products for its customers.Expanding presence at OCP Summit with next-generation server solutionsRamble Zhang, Associate Vice President of Global R&D and Product Development at Chenbro, highlighted that the Open Compute Project (OCP) was established in 2011 by Meta (formerly Facebook), Intel, Rackspace, and Goldman Sachs as a non-profit organization. Its mission is to promote open data center hardware architectures, drive product designs, and foster practical applications for its members and the industry.The initial aim of OCP was to provide ready-to-use hardware solutions for cloud service providers (CSPs) offering content or application services, significantly benefiting Meta due to its large-scale requirements. As a result, OCP has gradually evolved into a robust community organization, profoundly impacting the entire IT industry. Key technologies such as the OCP I/O module specification and OAM architecture have been widely adopted and recognized, with the OCP I/O module becoming a standard in data centers alongside NVIDIA HGX and OAM architectures. While other CSPs like AWS, Microsoft, and Oracle have not fully adopted OCP architectures, they have referenced relevant specifications, leading to many "OCP-like" product developments.According to Ramble, with the growth of standard OCP products and OCP-like products, OCP has become one of the most influential standard-setting organizations worldwide, spanning from upstream IC companies like Intel, AMD, and NVIDIA, as well as midstream CSPs, ODMs, SIs, and downstream component suppliers. This is one of the main reasons Chenbro decided to participate in the OCP summit this year. Previously, Chenbro primarily assisted its clients in attending and showcasing products. This year, however, the company aims to establish itself as a key player in the industry. In addition to showcasing its new generation of server chassis solutions at the OCP Summit, Chenbro is also planning to join the OCP Committee to make a more significant contribution to the IT industry.DC-MHS and MGX: Converging in the OCP ecosystemDC-MHS was one of the critical technologies Chenbro presented at the summit. This technology aims to establish open standards and build up a flexible and scalable ecosystem to meet modern data centers' growing demands in computing, transmission, and storage. The development of DC-MHS can be traced back to early standards like ATX and SSI, which addressed general-purpose requirements in a more efficient hardware environment.Since 2010, the SSI specification has struggled to meet the demands of high-performance data centers, prompting the industry to explore various proprietary specifications. However, as the industry's needs converged, more companies began to focus on open standards, ultimately forming the preliminary framework for DC-MHS in 2020.Chenbro collaborated with Intel to develop the first DC-MHS proof-of-concept (POC) platform and worked with numerous ODMs and SIs to launch various general-purpose computing products based on the platform. Subsequently, AMD also joined the DC-MHS ecosystem, and the involvement of Dell and HPE further accelerated the standard's development and adoption.The convergence of DC-MHS with the OCP standard has further propelled its growth, enabling the integration of CSP demands and ultimately leading to the creation of the new OCP standard, SDNO. Chenbro now fully supports SDNO and is actively collaborating with leading IC companies, OEM manufacturers, and system integrators to promote the adoption and development of this new standard.In addition, Chenbro's MGX products for AI applications are closely aligned with the OCP SDNO standard, forming a complementary relationship with DC-MHS. Whether it's DC-MHS or MGX, both continuously strengthen their connections with OCP. By leveraging this ecosystem platform, Chenbro is driving the diversification of data center solutions."While companies may start from different points, they often face similar challenges and issues, leading to similar solutions," Ramble explained. "These commonalities can be consolidated into an industry consensus, ultimately forming standardized solutions. This is why now is the best time to proactively promote key products based on the DC-MHS and MGX platforms, raising market awareness and bringing these solutions to a broader range of potential customers."Embracing robust architectures to create a structured and predictable futureThe development of DC-MHS is supported by two major specifications: FLW and DNO. FLW is primarily designed for dual-socket configurations to meet high-performance computing demands, while DNO focuses on single-socket designs aiming to support mainstream applications, especially in scenarios with multiple I/O requirements.The subsequent launch of SDNO inherits the architectures of FLW and DNO, incorporating more CSP requirements to provide greater flexibility. From DC-MHS to SDNO, both maintain a high degree of interoperability in terms of design philosophy and specifications. For instance, many well-known 1U and 2U general-purpose server products are based on the FLW architecture, featuring high-performance configurations such as "Dual Socket + 32 DIMMs." In contrast, DNO is about half the size of FLW but supports more I/O configurations within that space, making it particularly suitable for GPU application scenarios.Taking NVIDIA's MGX architecture as an example, the latest product, GB200, is designed based on the DNO architecture. MGX provides CSP customers with server applications in 1U and 2U sizes, such as the NVL72, launched earlier this year, deploying a 1U Compute Tray within a single cabinet, supporting up to 72 B200 chips. The NVL36 extends this to a 2U Compute Tray application. In enterprise applications, MGX offers a 2U standalone model that supports traditional PCIe GPU cards, with configurations of up to four GPUs and supports air-cooled solutions. MGX also offers a 4U model capable of accommodating 8 to 16 PCIe GPU cards, available with both air-cooled and liquid-cooled solutions. The MGX product line covers various configurations, from high-density 1U to 2U and 4U models, with different cooling modes to address multiple application needs.Ramble emphasized that whether it's DC-MHS or MGX, both are architectures around which Chenbro builds its ecosystems. These architectures support different application scenarios, and it is thanks to these standards and frameworks that the entire industry can develop in a structured and orderly manner. This year, Corona Chen, CEO of Chenbro personally attended the OCP Summit, showcasing the company's commitment to developing these ecosystems alongside upstream and downstream partners, providing better services for end customers, and guiding the future direction of technology.Chenbro showcases flexible, scalable server designs for AI and data centers at the 2024 OCP SummitPhoto: Company

In recent years, digital health has become a new development trend in the global health and medical industry. By using information and communication technology (ICT) alongside a range of wearable measurement devices and together with big data and artificial intelligence (AI) technology, it all combines to greatly enhance health care system efficiency. As an example in the health care arena, diabetes prevention and care measures are vital. From the International Diabetes Federation (IDF) 2021 report, approximately 537 million people worldwide suffer from diabetes, a figure which is expected to rise to 643 million by the year 2030. It is therefore no surprise that blood glucose meter products are exhibiting a growing trend. As a result of technological developments, in recent years, wearable blood glucose meters (CGM) are providing an effective way of monitoring blood sugar levels. This provides a means for users to enhance their diabetes management, resulting in rapid growth for such products. A research report by Allied Market Research shows that the glucose monitoring system market size was US$6.62 billion in 2021 and is expected to expand to US$31.7 billion by 2031. This shows a compound annual growth rate of 17%.Holtek has accumulated extensive experience in the health measurement market for products that measure blood sugar, blood pressure, body fat, body temperature, heart rate, blood oxygen, and other health parameters. Since 2020, the number of related devices is approaching 300 million. For traditional blood glucose meters (BGM), the company has released a range of precision measurement MCUs such as the BH67F2472 and the USB interface type BH67F2475, which are suitable for general blood glucose meters. The BH67F2476 and the USB interface type BH67F2495 are used in high-precision blood glucose meter products and have been widely utilized both in domestic and overseas markets. All of these devices are utilized by major health measurement product manufacturers with more than 100 million blood glucose meter devices being sold since 2020. Holtek has also launched complete device solutions that incorporate the advantages of small size, low power consumption, and high precision for continuous glucose meters (CGM). These include the BLE MCU HT32F67595 and the CGM blood glucose meter AFE MCU, the BH66F2475.Comparison between Holtek AFE MCU and RF MCUHoltek CGM System SolutionPower Supply: uses CR1216 button cell, continuous use for 24 daysSize: 23mm diameterAccuracy: ±0.05nA when sensor current is between 0 and 30nA.Holtek CGM DataThrough its solution company, BestHealth, Holtek provides an extensive range of development support for its health measurement product customers. The company provides technical services, which include product development platforms, complete solutions, and professional modules. These core technologies are aimed at minimally invasive blood sampling analysis (blood sugar, uric acid, blood ketones, total cholesterol), HCT analysis (multi-frequency AC impedance/phase angle), CGM continuous dynamic blood glucose monitoring technology, electronic sphygmomanometer technology, PPG measurement technology (blood oxygen, heart rate, blood pressure, HRV heart rate variability), PI (perfusion index), transmission blood oxygen measurement technology (heart rate, blood oxygen), thermopile infrared temperature measurement technology, etc.