Going Beyond the Limits with Advanced Heterogeneous Integration

Advanced heterogeneous integration is a method of streamlining and improving the process of integrating separately manufactured chip components into a single assembly level. Arguably key to the refinement of the chip production industry, companies that master advanced heterogeneous integration can significantly improve the efficiency of production processes and performance and do more with less. Samsung Foundry aims to do just that, as explained by Moonsoo Kang, Head of Business Development at Samsung Foundry, at the 2022 Samsung Foundry Forum (SFF). Kang used his presentation at the forum to lay out in detail the company’s roadmap for its advanced heterogeneous integration solutions. With a message rooted in Samsung’s ‘Beyond Moore’ approach, Kang also gave visitors a deep dive into the advanced package technologies that are powering new innovations in horizontal and vertical integration: I-Cube, H-Cube and X-Cube. Prior Approaches: ‘More Moore’ and ‘More Than Moore’ At the opening of his talk, Kang revisited the integration approaches that dominated previous eras of chip development, namely the ‘More Moore’ and ‘More Than Moore’ approaches. Both are named in reference to engineer Gordon Moore’s 1965 prediction that every two years would see a doubling of the transistor count on chips, which is also known as ‘Moore’s Law.’ The ‘More Moore’ approach, Kang explained, sought to relentlessly scale down transistors and interconnects on silicon and, by doing so, enabled a revolution in computing and electronics. However, as demand for computing power has skyrocketed – necessitating more transistors in the process – the ‘More Moore’ approach reached its limits. Fitting as many functions as possible within a single chip can be economical and practical. However, as the industry has reached the maximum reticle size that can be printed at once using photolithography methods – which use light to add materials onto substrates – scaling alone cannot be relied on as the sole solution for high performing chips. The ‘More Than Moore’ approach, by contrast, seeks to create specialized and diverse chip architectures to best suit the wide array of functions required. Effective, certainly, but not always economical. Despite their apparent differences, ‘More Moore’ and ‘More Than Moore’ have, Kang argued, complimentary potential. With this in mind, Kang and the teams at Samsung Foundry have been pursuing a third way: ‘Beyond Moore.’ This approach seeks to achieve the best of both worlds, and the key to doing so, Kang explained, is advanced heterogeneous integration. ‘Beyond Moore’ and the Role of Advanced Heterogeneous Integration

Through ‘Beyond Moore,’ heterogeneous integration can mitigate increases in development costs and boost performance. “As more features are added onto a chip, the die sizes increase, despite good Moore-like scaling,” Kang said. “By partitioning a large die into smaller chiplets and adopting an optimal process for each chiplet, we hope to improve overall yield and reduce manufacturing costs.” Customers have also explored the benefits of reusing chiplets. Reusing most of the IP in a product as chiplets, redesigning only the excluded area and then combining it all using heterogeneous integration techniques can result in significant time and cost reductions. The scope for performance improvement is being investigated too. “With conventional 2D designs, signal paths between design blocks can be several millimeters long,” Kang explained. “However, with 3D integration, one chip sits directly on top of another. So, they can be cut down to a few micrometers, leading to a major drop in latency. Plus, finer interconnect pitch in 3D integration enables extremely high bandwidth, which leads to higher performance.” Three Cubes for the Future of Foundry Kang went on to discuss the technologies that have resulted from the ‘Beyond Moore’ approach and foundry experiments in heterogeneous integration. These advanced package technologies – or ‘Cubes’ – come in three varieties based on two axes – horizontal and vertical integration: I-Cube, H-Cube and X-Cube.

I-Cube The I-Cube is a 2.5D solution, which places chips horizontally on top of an interposer. Kang explained that I-Cube customers are demanding larger interposer sizes that deliver them the higher computing performance they need. To address this, Samsung offers two types of I-Cube: I-CubeS and I-CubeE. “As the latest I-CubeS advancement, I-CubeS 8 has an interposer made of silicon and features a 3x reticle size to carry eight HBMs and two logic dies,” Kang elaborated. I-CubeE, meanwhile, stands for Embedded Silicon Bridge die, and is a more cost-effective option for large interposers. “We also plan to expand the size of I-CubeE so that it can integrate up to 12 HBMs for mass production in 2025.” H-Cube H-Cube – another 2.5D solution – stands for Hybrid Cube. This solution was designed to respond to a severe shortage in unit printed circuit boards (PCBs) currently facing the industry. “By converting a large ABF substrate into a combination of a smaller ABF substrate or FBGA substrate and a larger HDI substrate, we are able to provide our customers with advanced PCB solutions with excellent characteristics, lower PKG cost and easier PCB SCM,” Kang explained. X-Cube X-Cube is a full 3D solution, which allows the vertical stacking of chips. “X-Cube comes in two different forms: Two vertically stacked dies are connected either with micro-bumps, or by bump-less Cu-to-Cu direct bonding,” said Kang. “We plan to start mass production of X-Cube products with micro-bump type in 2024, and bump-less type in 2026.” Kang laid out roadmaps for the expansion and further development of all the company’s Cube solutions, some of which will see new commercial versions this year. “Samsung Foundry’s Cube will offer flexible architecture to meet a variety of customer needs and applications,” he said. Other Advancements and the Growth of a Seamless Ecosystem

Beyond Cube solutions, Kang detailed the other technologies and advancements that Samsung Foundry has been focusing on, Integrated Stack Capacitor (ISC) developments and close collaboration with the Samsung Memory Division. This collaboration is focused on a new product: direct 3D stacked DRAM on a logic die, without buffer die, to support the next generation of high-performance computing (HPC) devices. Kang’s final subject was one he believed to be very important to the business: ecosystem. “We at Samsung Foundry, have worked very hard to build a strong foundry ecosystem with our partners,” Kang emphasized. “Our ecosystem for ‘Beyond Moore’ today includes a large number of partners, such as Design Solution Partners (DSP), IP vendors, EDA partners, OSATs and even PCB suppliers. By closely collaborating with our eco partners, we will accommodate a host of different business models and customer needs and provide seamless support in a highly flexible way.” “We in the semiconductor industry are currently facing many challenges and limitations. But, at the same time, this is a very exciting moment with many changes and innovations to come. Samsung Foundry is ready to be your trusted and valued partner on this exciting journey to a new frontier,” he concluded.