News
Driven by the AI chip wave, “advanced packaging” emerges as the hottest technology in the semiconductor industry. Its significance extends beyond computational power demands, as the escalating cost of semiconductor processes and the limits of Moore’s Law make the “integration capability” of advanced packaging a crucial weapon for industry players to break through.
According to a report from TechNews, TSMC, Intel, and Samsung have all been deeply involved in advanced packaging for many years and have already introduced corresponding solutions. However, these semiconductor giants are not only focused on this aspect.
In addition to their own technologies, they are actively fostering supply chains, setting standards, and building ecosystems. By accelerating the development of advanced packaging technology, they are also laying the groundwork for their future influence.
Intel, for instance, has chosen to start with standardization by proposing the Universal Chiplet Interconnect Express (UCIe) alliance. Through open specifications and standardized connections, the protocol directly adopts mature standards like PCI Express (PCIe) and the recently developed Compute Express Link (CXL).
The reason for starting with chiplet technology is that in recent years, more and more semiconductor companies have discovered that designing chips using Chiplet architecture and integrating them through advanced packaging technology is more cost-effective than traditional System-on-Chip (SoC) approaches.
Therefore, Intel’s focus on connecting chiplets through standards like UCIe is aimed at providing a standardized interface stack for complete chiplet integration. UCIe supports 2D, 2.5D, and bridge packaging, with future development expected to include support for 3D packaging as well.
Intel’s Packaging Test Technology Development Department’s Senior Chief Engineer, Zhiguo Qian, directly involved in the UCIe Alliance, emphasizes that advanced packaging has become a crucial aspect of semiconductor development, particularly in ensuring the continuation of Moore’s Law.
Qian further points out that when considering the impact of the UCIe standard on the advanced packaging industry, it indeed establishes a standard for interconnecting chiplets within SoCs. This was the original intent behind Intel’s promotion of the UCIe standard alliance.
Currently, advanced packaging is mostly divided into different structures like 2.5D and 3D, and some even classify it as 2.1D or 2.2D, showcasing diverse structural designs across the industry.
However, within these structures, each company has its own proprietary interface solutions, and some even offer multiple solutions. Therefore, to meet customer demands, these standard interconnections must not only be at the forefront of technology but also be compatible with various standards that are open and do not incur any licensing fees.
On the other hand, the UCIe alliance has established various standards, such as the required packaging architectures and interface wiring designs, to achieve the desired performance levels. These standards provide guidelines for customers seeking advanced packaging solutions. By adhering to UCIe standards, customers can anticipate the performance of their chips, without the need for trial and error(in the IC designing stage).
Currently, companies participating in the UCIe alliance include Qualcomm, AMD, Arm, NVIDIA, TSMC, ASE Group, Winbond Electronics, and Applied Materials, among others, along with semiconductor giants like Samsung. Additionally, Google Cloud, Microsoft, and Meta are members, alongside over 120 other companies.
TSMC is also focused on ecosystem development, as evidenced by its announcement of the 3DFabric Alliance within the Open Innovation Platform (OIP) during the 2022 Open Innovation Platform Ecosystem Forum.
In fact, the 3DFabric Alliance is built upon TSMC’s 3DFabric technology introduced in 2020. This technology encompasses a comprehensive solution ranging from advanced processes to silicon stacking and advanced packaging technologies such as CoWoS and InFO.
With an established customer base for its 3DFabric technology, TSMC expanded it into an alliance in 2022. The goal is to assist customers in achieving rapid implementation of chip and system-level innovations while strengthening TSMC’s influence in advanced packaging.
The 3DFabric Alliance marks TSMC’s sixth open innovation platform alliance and is the semiconductor industry’s first alliance aimed at accelerating innovation and enhancing the 3D Integrated Circuit (3D IC) ecosystem in collaboration with partners.
This alliance includes companies in electronic design automation (EDA), silicon intellectual property (IP), design center alliances (DCA)/value chain alliances (VCA), memory, outsourced packaging testing (OSAT), and substrate and testing. Members include Ansys, Cadence, Siemens, ARM, Micron, Samsung, SK Hynix, Amkor, ASE, Advantest, and more.
In addition to establishing the alliance, TSMC also introduced the 3Dblox standard during the alliance’s inception. This standard integrates the design ecosystem with validated EDA tools and processes to support 3DFabric technology.
The purpose of this standard is to break the complexity of 3D IC design caused by each EDA supplier using its preferred language. Through the modular 3Dblox standard, key physical stacking and logic connection information in 3D IC design are standardized in a single format, simplifying input and significantly enhancing interoperability among different tools in 3D IC design.
From Intel’s UCIe standard to TSMC’s 3DFabric alliance and 3Dblox standard, it’s evident that in the era of advanced packaging, the key to solidifying the positions and market shares of semiconductor giants lies not only in their individual technological breakthroughs but also in their ability to coordinate and integrate the upstream and downstream industries.
Read more
(Photo credit: TSMC)
Insights
The semiconductor industry enters the era of integration. Various foundries are focusing on advanced packaging technologies, but the terminology surrounding advanced packaging can be daunting. This article aims to explain these terms in the simplest way possible.
According to a report from TechNews, currently, there are two main trends in advanced packaging: heterogeneous integration and chiplets.
In fact, the concept of “heterogeneous integration” has been developing for many years and is not exclusive to advanced packaging. It is not only used for the integration of heterogeneous chiplets but also for integrating other non-chip active/passive components into a single package, which is the technology commonly used in traditional Outsourced Semiconductor Assembly and Test Services(OSATs).
In the simplest terms, “heterogeneous integration” can be likened to building with large building blocks, while “advanced packaging” is akin to assembling with small building blocks. Some manufacturers, like traditional Outsourced Semiconductor Assembly and Test Services(OSATs), excel in stacking large blocks, such as logic circuits, radio frequency circuits, MEMS (Micro-Electro-Mechanical Systems), or sensors, onto a IC substrate. The stacking of these different large blocks represents the concept of heterogeneous integration.
On the other hand, some blocks are too small to stack effectively, requiring assistance from advanced packaging, typically provided by semiconductor foundries.
Advanced packaging also encompasses 2.5D packaging and 3D packaging. Using the metaphor of building blocks, the former involves horizontally stacking small building blocks on a interposer, while the latter involves vertically stacking small building blocks with interconnection facilitated through Through-Silicon Vias (TSVs), which are ultra-small building blocks.
It’s important to emphasize that stacking blocks is a conceptual representation, and the distinction between large and small blocks is relative. The analogy above refers to heterogeneous integration in traditional packaging, and heterogeneous integration in advanced packaging follows a similar concept, but with even smaller building blocks.
With this concept in mind, let’s discuss the applications of heterogeneous integration in advanced packaging:
Among the various packaging types, SoC (System On Chip) involves integrating different chips such as processors and memory, with different functions, redesigned and fabricated using the “same process,” integrated onto a single chip, resulting in a final product with only one chip.
On the other hand, SiP (System in Package) involves connecting multiple chips with “different processes” through “heterogeneous integration” technology, integrated within the same packaging module. Therefore, the final product will be a system with many chips on it, resembling the stacking of different-sized building blocks mentioned earlier.
Therefore, heterogeneous integration refers to integrating different and separately manufactured components (heterogeneous) into higher-level assemblies. These components include blocks of different sizes, such as MEMS devices, passive components, logic chips, and more.
However, at a certain point, for the sake of process development, researchers found that separating components at the right time might facilitate miniaturization. Hence, chiplet was born.
As demands for ICs become increasingly complex, the size of SoC chips continues to grow. However, cramming too many components onto a limited substrate poses significant challenges, including heightened process complexity and reduced yield.
Hence, the concept of chiplets emerged, advocating for the segmentation of SoC functionalities, such as data storage, computation, signal processing, and data flow management, into smaller individual chips. These chiplets are then integrated through packaging to form a interconnected network.
It’s worth noting that Chiplets are essentially chips, whereas SiP refers to the packaging format. Chiplet architecture enable the reduction of individual chip sizes, simplify circuit design, overcome manufacturing difficulties and yield issues, and offer greater design flexibility.
Among them, there are two integration methods for the chiplet mode: “Homogeneous Integration” and “Heterogeneous Integration”. In many cases, both integrations actually coexist.
Homogeneous Integration involves designing two or more chips and then using advanced chip integration techniques to combine them into a single chip. On the other hand, heterogeneous integration of chiplets involves integrating different types of logic chips, memory chips, etc., using advanced packaging techniques because different types of chips cannot be manufactured in the same process.
For example, Apple and TSMC’s collaboration on custom packaging technology, UltraFusion, connecting two M2 Max chips to introduce the M2 Ultra, falls under the category of homogeneous chiplet mode. At the same time, integrating CPU, AI accelerators, and memory into AI chips belongs to the heterogeneous mode, such as AMD’s launch of CCD (Core Chiplet Die) chiplet products in 2020, enhancing design flexibility.
Currently, advanced packaging can be broadly categorized into three main types: Wafer-Level Packaging (WLP), 2.5D Packaging, and 3D Packaging. Traditional packaging involves cutting wafers into chips before packaging, while advanced packaging entails packaging the silicon wafer before cutting, requiring subsequent stacking processes in fabs. Therefore, the technology is primarily the responsibility of fabs.
Traditional packaging involves cutting wafers into chips before packaging. Advanced packaging, starting from wafer-level packaging, involves packaging silicon wafers before cutting, and subsequent stacking requires wafer fabrication processes.
Therefore, this article will delve into advanced packaging technologies offered by the three major foundries, with a focus on 2.5D and 3D packaging.
To further explain using building blocks, the difference between 2.5D and 3DIC packaging lies in the “stacking method.”
In 2.5D packaging, processors, memory, or other chips are stacked horizontally on a silicon interposer using a flip-chip method, with micro bumps connecting different chip’s electronic signals. Through silicon vias (TSVs) in the interposer link to the metal bumps below, then packaged onto the IC substrate, creating tighter interconnections between the chips and the substrate.
In a side view, although the chips are stacked, the essence remains horizontal packaging, with the chips positioned closer together and allowing for smaller chip sizes. Additionally, this is a form of “heterogeneous integration” technology.
3D packaging involves stacking multiple chips (face down) together, directly using through-silicon vias to stack them vertically, linking the electronic signals of different chips above and below, achieving true vertical packaging. Currently, more and more CPUs, GPUs, and memories are starting to adopt 3D packaging technology.
Hybrid bonding is one of the die bonding techniques used in advanced chip packaging processes. One of the commercially available technologies in this domain is the “Cu-Cu hybrid bonding.”
In traditional wafer bonding processes, there are interfaces between copper and dielectric materials. With “Cu-Cu hybrid bonding,” metal contacts are embedded within the dielectric material. Through a thermal treatment process, these two materials are bonded together, utilizing the atomic diffusion of copper metal in its solid-state to achieve the bond. This approach addresses challenges encountered in previous flip-chip bonding process.
Compared to flip-chip bonding, hybrid bonding offers several advantages. It allows for achieving ultra-high I/O counts and longer interconnect lengths. By using dielectric material for bonding instead of bottom fillers, the cost of filling is eliminated.
Additionally, hybrid bonding results in minimal thickness compared to chip-on-wafer bonding. This is particularly beneficial for future developments in 3D packaging, where stacking multiple layers of chips is required, as hybrid bonding can significantly reduce the overall thickness.
As the semiconductor industry enters the “post-Moore’s Law era,” the development focus of advanced packaging is gradually shifting from 2D planar structures to 3D stacking and from single-chip designs to multi-chip configurations. Therefore, “heterogeneous integration” will play a crucial role in future advanced packaging.
Currently, prominent companies such as TSMC, Samsung, and Intel are intensifying their research and development efforts and capacity expansions in this field, introducing their innovative packaging solutions.
With ongoing technological advancements and innovations, advanced packaging and heterogeneous integration will play increasingly vital roles in propelling the semiconductor industry towards greater heights, meeting the complex and diverse demands of future electronic devices.
Read more
(Photo credit: Intel)
News
In 2022, Intel engaged in negotiations with the Italian government, planning to invest USD 5 billion in constructing a packaging and testing facility. This project would also receive subsidies from the Italian government, expected to cover 40% of the construction costs, along with additional subsidies or incentives. Furthermore, Intel intended to establish a research and design center in France, expected to create a complete semiconductor supply chain in Europe.
However, according to a report from Reuters, Italian Minister of Industry Adolfo Urso has indicated that Intel may delay or abandon its investment plans in Italy and France to fulfill its prior commitments in Germany. Nonetheless, Italy has not completely given up on attracting Intel; Adolfo Urso emphasizes that Italy remains very welcoming if Intel changes its mind.
On the other hand, according to another previous report from Reuters, the US government is expected to announce a significant grant for Intel’s Arizona project soon. This grant will be part of the USD 39 billion direct appropriations and USD 75 billion loans and guarantees under the “Chip Act.”
Among the recipients of subsidies under the “Chip Act,” Intel is expected to receive the largest subsidy to date. According to a previous report from Tom’s Hardware, Intel is anticipated to receive a government subsidy of USD 10 billion, with TSMC and Samsung potentially included in the latest subsidy list as well.
Samsung Electronics is, according to its own expectation, investing USD 17 billion to construct a foundry in Taylor, Texas, while TSMC is investing roughly USD 40 billion to build a foundry in Phoenix, Arizona. However, there are rumors suggesting that due to the U.S. prioritizing domestic companies, the expected subsidy amounts for Intel may differ from those for TSMC and Samsung.
The U.S. government enacted the “Chip Act” in 2022, but subsidies have been modest, with only three American companies currently benefiting, including BAE Systems, GlobalFoundries, and Microchip Technology.
In order to accelerate the development of the IDM 2.0 initiative, Intel made a significant expansion decision in 2021, investing approximately USD 20 billion in the Octillo campus in Arizona, USA. This investment involved the construction of two new fabs and the implementation of EUV production lines to support Intel’s 20A and Intel’s 18A process technologies. The new Fab 52 and Fab 62 are expected to commence operations in 2024.
Read more
(Photo credit: Intel)
News
Chip giant NVIDIA kicked off its annual Graphics Processing Unit (GPU) Technology Conference (GTC) today, with CEO Jensen Huang announcing the launch of the new artificial intelligence chip, Blackwell B200.
According to a report from TechNews, this new architecture, Blackwell, boasts a massive GPU volume, crafted using TSMC’s 4-nanometer (4NP) process technology, integrating two independently manufactured dies, totaling 208 billion transistors. These dies are then bound together like a zipper through the NVLink 5.0 interface.
NVIDIA utilizes a 10 TB/sec NVLink 5.0 to connect the two dies, officially termed NV-HBI interface. The NVLink 5.0 interface of the Blackwell complex provides 1.8 TB/sec bandwidth, doubling the speed of the NVLink 4.0 interface on the previous generation Hopper architecture GPU.
As per a report from Tom’s Hardware, the AI computing performance of a single B200 GPU can reach 20 petaflops, whereas the previous generation H100 offered a maximum of only 4 petaflops of AI computing performance. The B200 will also be paired with 192GB of HBM3e memory, providing up to 8 TB/s of bandwidth.
NVIDIA’s HBM supplier, South Korean chipmaker SK Hynix, also issued a press release today announcing the commencement of mass production of its high-performance DRAM new product, HBM3e, with shipments set to begin at the end of March.
Recently, global tech companies have been heavily investing in AI, leading to increasing demands for AI chip performance. SK Hynix points out that HBM3e is the optimal product to meet these demands. As memory operations for AI are extremely fast, efficient heat dissipation is crucial. HBM3e incorporates the latest Advanced MR-MUF technology for heat dissipation control, resulting in a 10% improvement in cooling performance compared to the previous generation.
Per SK Hynix’s press release, Sungsoo Ryu, the head of HBM Business at SK Hynix, said that mass production of HBM3e has completed the company’s lineup of industry-leading AI memory products.
“With the success story of the HBM business and the strong partnership with customers that it has built for years, SK hynix will cement its position as the total AI memory provider,” he stated.
Read more
(Photo credit: NVIDIA)
News
Benefited from robust demand in downstream application markets, the silicon carbide (SiC) industry is in high gear. According to TrendForce, the SiC power device market is expected to reach USD 5.33 billion by 2026, with its mainstream applications still highly reliant on electric vehicles and renewable energy sources.
Recently, the widely-publicized SiC market has seen new developments involving companies such as Mitsubishi Electric, Mersen, and Ascen Power.
According to recent reports from Nikkei, Mitsubishi Electric plans to commence construction of a new 8-inch SiC fab in Kumamoto Prefecture, Japan, in April 2024, with operations scheduled to start in April 2026.
In March 2023, Mitsubishi Electric announced the plan to invest approximately JPY 100 billion (Around CNY 4.856 billion) over five years to construct an 8-inch SiC fab and enhance related production facilities. The fab is projected to kick-start operation in April 2026.
The new fab, spanning six floors with a total floor area of around 42,000 square meters, will primarily handle front-end processes for 8-inch SiC wafers. Mitsubishi Electric will introduce an automated transport system across all processes to create a highly efficient production line and plans to gradually increase capacity, aiming to increase SiC production capacity by five times by the fiscal year 2026 (compared to fiscal year 2022).
In May 2023, Mitsubishi Electric signed a MOU with Coherent to supply 8-inch n-type 4HSiC wafers for the new factory. Both parties are committed to expanding the production scale of 8-inch SiC devices.
Recently, Shao Yonghua, the plant manager of Ascen Power’s fab, introduced that the entire fab is currently ramping up capacity, with the planned capacity of producing 240,000 pieces of 6-inch automotive-grade SiC chips annually expected to be achieved by the end of this year.
The reserved 8-inch production line is adjacent to the 6-inch line and will have the capability to produce 240,000 pieces of 8-inch automotive-grade SiC chips annually once completed.
As previously reported, Ascen Power’s SiC chip manufacturing project is a major project under Guangdong’s “Strengthening Chip Technology Project,” with a total investment of CNY 7.5 billion, covering an area of 150 acres.
The first phase involves an investment of CNY 3.5 billion to build a production line capable of producing 240,000 pieces of 6-inch SiC chips annually, with the second phase focusing on establishing a production line capable of producing 240,000 pieces of 8-inch SiC chips annually. The products include IGBTs, SiC SBD/JBS, SiC MOSFETs, targeting applications including new energy vehicles, photovoltaics, smart grids.
In November 2022, the project’s clean room was officially put into use, achieving a monthly production capacity of 10,000 pieces. Its automotive-grade and industrial-grade chips have been successfully mass-produced and sampled, and these chips are about to complete the automotive verification. Up to now, Ascen Power has signed agreements with more than 40 customers and achieved tape-out, covering most SiC chip design companies nationwide.
On March 12, European graphite materials and silicon carbide wafer supplier Mersen announced that it has received investment from the French government for capacity expansion of its SiC wafer project. The subsidy amount may exceed Euro 12 million (Approximately CNY 94 million), sourced from the “France 2023 Plan”—a significant joint interest project in microelectronics and communication technology in Europe.
Mersen stated that they intend to advance the research and industrial production of p-SiC wafers with this investment. p-SiC is a low-resistivity polycrystalline SiC wafer that can be combined with single-crystal SiC active layers, enabling SiC device manufacturers to improve production yield and transistor performance.
Mersen expects to invest Euro 85 million (Approximately CNY 670 million) between 2023 and 2025, employ 80 to 100 staff, promote capacity construction at the Gennevilliers plant in France, and accomplish a potential manufacturing capacity of 400,000 wafers (150mm) by 2027.
Additionally, Mersen will supply SiC wafers to Soitec. In November 2021, two sides entered into a strategic partnership to jointly develop polycrystalline SiC wafers with extremely low resistivity for SiC power electronic components based on Soitec SmartSiC technology, leveraging their respective expertise in substrates and materials.
Read more
(Photo credit: Mitsubishi Electric)