My perspective on building a successful systems foundry

The digitization of everything is accelerating the global demand for semiconductors. And the industry is forging ahead to meet this demand. I have been part of Intel for almost 30 years now, and I am more excited than ever as we unleash our silicon design and manufacturing expertise to build world-changing products for our customers.

In this article, I share my insights on building a successful systems foundry with the emergence of the new open, collaborative semiconductor design and manufacturing ecosystem.

Ecosystem partners are critical

The first thing I have learned is that a foundry never thrives by itself. It takes a village, so to speak, to raise a semiconductor product. There are ecosystem partners who build intellectual property (IPs), system integrators who create the designs, outsourced semiconductor assembly and test companies (OSATs) that we can leverage for the package, electronic design automation (EDA) tool vendors, and so on. Only with this entire ecosystem around it, a foundry can be successful.

The global semiconductor IP market is predicted to grow to around $12 billion by 2031, growing at a CAGR of ~6% over the forecast period.?EDA market size is projected to cross $21 billion by 2028 from about $11 billion in 2020, growing at a?CAGR of 9%.

To put it in simple terms, a foundry that works alone, creating only custom designs for a niche clientele, is like a couture fashion house that offers unique, made-to-measure garments. To thrive, a foundry needs to bring the ecosystem along and be more like a department store, offering ready-to-wear apparel in standard sizes for everyone.

For example, early this year, Intel launched IFS (Intel Foundry Services) Accelerator, a comprehensive ecosystem alliance to help foundry customers smoothly bring their silicon products from idea to implementation. Through this alliance, which includes 17 founding partner companies, chip designers will have access to full support from EDA and design services providers and a broad library of IP offerings from a range of partners.

IP: Semiconductor IP cores are reusable blocks of logic that implement a major function. With the increasing complexity of system-on-chip (SoC) design, IP-based design has become the preferred path to achieve high-performance designs and accelerate time-to-market. It’s important to partner with industry-leading IP providers to enable designers to access high-quality IPs, supporting their design needs and project schedule while optimizing for performance, power, and area (PPA). The IP portfolios include all the essential IP blocks needed for modern SoCs – a comprehensive, silicon-verified, and Intel process-specific suite comprising standard cell libraries, embedded memories, general purpose I/Os, analog IP, and interface IP.

Design IP market is soaring; interface IP is on a surge: The industry is seeing significant traction in the design IP market due to the growth in data-centric applications, networking, automotive, and AI. According to WSTS, design IP sales reached $5.45B in 2021, registering 19.4% YoY growth. Moreover, it has recorded almost 10% CAGR for 2016-2021.

The broad categories within design IP market are processor (CPU, DSP and GPU IP), interface (integration of the protocol-based functions like USB, PCI Express, Ethernet, MIPI, SATA, DP, Die-to-Die or D2D interface and memory controller - DDRn, LPDDRn, HBM, GDDR), other physical IP (aggregation of various categories like SRAM memory compilers, physical library, analog & mixed signal, wireless interface etc.) and digital IP.

Interestingly, the market for interface IP is showing strong growth and is estimated to rise to $2.5 billion in 2025, driven by growth in PCIe, DRR and ethernet & D2D. The trend in favor of interface IP is clear. The market share of processor IP (CPU, GPU and DSP) has moved down from 53.5% in 2018 to 51% in 2020, while interface IP category has risen from 20.3% to 23.2 % during the same period. In 2016 processor IP was 63.8% of the market while interface IP constituted only 16.9%. The key drivers are data-centric application, hyperscalar, datacenter, storage, wired and wireless networking and AI. All these applications share the need for higher and higher bandwidth, driving the growth for interface protocols like PCIe, Ethernet and SerDes or memory controller IP.

Another important aspect is that as we advance to the latest process nodes, analog components of the interface IPs continue to get more complex. These designs are increasingly susceptible to process variations and hard to scale and port with each advancing process node. A lot of innovations are underway – analog as well as DSP-based. These DSP-based approaches help fine-tune the signal receiver in the field for PVT variations.

As we develop the systems foundry, collaborating with key players is important.

EDA: The most powerful and validated EDA solutions optimized for Intel's leadership technology and manufacturing, covering the full spectrum from concept to high-volume silicon production. EDA suppliers produce tools that assist in the specification, planning, design, verification, implementation and testing of electronic systems. Collaborating closely with EDA partners allows customers to co-optimize and enhance tools and flows so chip designers can best realize their performance, power and area (PPA) goals, while accelerating their time to market.

Design services: Designing the next generation of semiconductor products requires skilled engineering talent and resources, especially when working with leading-edge process technologies. Partnering with design service providers gives customers options for additional support when bringing their ideas to life, with specialties ranging from analog and digital physical designs to low-level system software. Depending on a customer's needs, the silicon experts from the alliance partners can assist in different stages, including design, verification, implementation, and emulation.

Modularity and standards will enable chiplet-based architecture for semiconductor growth

In my previous article , I wrote about the advantages of modular, chiplet-based semiconductor architecture and the importance of industry standards to enable the ecosystem. The rationale for modularity is clear. There was a time when we were on 500 nm transistors. Now, we have 2 nm (20 Angstrom) and smaller on Intel’s process technology roadmap. Given the complexity and cost of these advanced nodes, chiplets offer another way to add functionality.

A chiplet needs to be connected with other chiplets, usually through advanced package integration such as EMIB (Embedded Multi-die Interconnect Bridge) and the use of standardized interfaces. That's why Intel developed UCIe (Universal Chiplet Interconnect Express), an open standards body that defines the interconnect between chiplets within a package. Over 80 companies have joined the UCIe consortium, coming together to create unified standards for interoperable chiplets.

Systems foundry is a paradigm shift to enable customer innovation

While a traditional foundry is a semiconductor manufacturer that makes chips for other companies, IFS is a systems foundry that offers silicon, packaging, software and chiplets. Intel leverages what we call an 'anchor customer methodology'. We give our anchor customers extra benefits, providing them with collaterals and Process Design Kits (PDKs) early on, so that they become our partners and we build the solution together. Therefore, building trust and transparency is absolutely critical. Our foundry model leverages the ecosystem for IP, design services, tools, flows and methodologies, and Intel comes with our strength in transistor design. So, from transistor development to IP development, design capability, test and validation, package, board – all of it can be supported.

Intel offers a menu of options to accelerate customers' designs. As we transition from a system on chip to a system of chips, we need to optimize everything from the system and software architecture to how we do manufacturing and test. We need to start with applications and system software to maximize the specific and aggregated performance.?We need to think about everything in 3D – from die construction to assembly rules, to verification and signoff, to our automated post-silicon verification and test flows.

With the systems foundry model, IFS will offer a secure supply, open platforms, and composable solutions through semiconductor chips, advanced packaging, and system software.

Customer-obsessed, solution mindset is the way forward

With this new systems foundry approach, the customer is king. This heralds a culture change for us. Historically, Intel has been focused on advancing process technology. We have focused on building cutting-edge performance on the leading nodes and offering our products to the ecosystem. Now, the mindset of a foundry is entirely different. We are listening to our customers and enabling the product that they need. Solutions are the way to go – the chip is a component, and we must work with ecosystem partners to build an optimal solution for our customer's success.

Intel has also committed capacity in our fabs for foundry customers. Our first large-scale foundry operation in Arizona, USA, will not only meet requirements for Intel products but also provide committed capacity for IFS customers.

Adopting a service mindset means for example, as a foundry, we need to support mature technology nodes, which currently account for half of the world's semiconductor requirements. We will not retire our process technology nodes but continue to support what our customers need and ask for. For example, automotive customers are looking for support over a lifecycle of 20+ years and we will provide that support, be it validation, debug or application engineering, etc.?

Besides Intel's leading-edge nodes, industries like the automotive sector, for instance, need differentiated specialty technologies on mature nodes. Given my role in IFS as the general manager of the automotive solutions group, I'm very excited about the revolution that the automotive industry is going through and how we can help our OEM customers as they go through the transition from internal combustion engines to hybrid and fully electric vehicles and the adoption of advanced driver assistance systems (ADAS) and autonomous driving (AD).

Three key takeaways from my learnings

What I have learned is that we must build products to meet the needs of the customer. Therefore, it is important to listen to what our customers are asking for. We need to provide them with solutions rather than components and extend lifecycle support. Secondly, not everything needs to be built in-house. We must build the ecosystem, offer an open platform, drive opensource and industry standards, and enable others to innovate. Lastly, modularity and scalability can be enabled through the systems foundry model that we are building with the broadest portfolio to support our customers' success. We are focusing on these few things to be a successful open systems foundry.