A growing RISC in semiconductors

Semiconductors's strategic importance can't be underestimated   

Semiconductors are already ubiquitous, yet this is only the beginning

Computing components are the bedrock of modern economies. Thanks to constantly increasing capabilities and decreasing costs (a consequence of the famous Moore’s Law), they have allowed the ongoing digitalization of our lives and have become present in virtually everything: more than 1tn semiconductor units are produced each year powering devices ranging from toys to critical infrastructure, from lighting to military vehicles.

They are also at the core of one of the next major revolutions experienced by mankind: the advent of artificial intelligence (AI), which demands a massive amount of computing power. In addition to its obvious strategic importance, AI represents a nearly $200bn projected opportunity by 2030 for electronic chips, with an expected ~37% CAGR through the decade.

Computing architectures are the keys of the kingdom

The way computing components operate is determined by their architecture, i.e., a set of instructions (ISA) representing all the basic operations supported by the chips which implement the architecture. The ISA, therefore, determines what the chips will be capable of, like the cogs of an industrial machine will determine the extent of its capabilities. Controlling the building blocks of an architecture means controlling its raison-d’être and what it is capable of. Consequently, controlling the destiny of the whole ecosystem built upon it, which can be truly massive.

Control is increasingly tightening

Over 40 ISAs have been created over time, yet today only 2 are produced at significant volumes: x86 and ARM. Only three companies have control over these architectures: Intel and AMD for the former, and ARM for the latter. x86 has a monopoly in high-performance applications (e.g., datacenters), while ARM dominates embedded applications (e.g., smartphones).  

The Intel/AMD dominance in x86 is problematic, but competitive dynamics remain quite healthy; this was demonstrated by the recent revival of AMD at the expense of Intel, which led to technological improvement at a better pricing for customers. On the other hand, ARM today wields an oversized power over the industry, as there is strictly no alternative: as highlighted by the tentative acquisition by semiconductor behemoth Nvidia, it would be "straightforward" to restrict access to the company's technology. While the acquisition was ultimately unfruitful due to regulatory scrutiny, this move acted as a wake-up call for many: it is critical to find a third way.  

RISC-V: an architecture available for free  

When open source meets hardware

RISC-V is an architecture created in 2010 at Berkeley University and originally developed for academic purposes. The core instruction set is open source, meaning that everybody can use it for free (like Linux in software), while you have to buy processors to AMD and Intel or to pay royalties to ARM to use their blueprints. A commercial license is however available for companies wishing to develop and sell their original designs.

More importantly, the instruction set has been designed with modularity in mind, meaning that custom instructions can be easily added to perform specific tasks if needed while maintaining compatibility with the rest of the ecosystem. Doing the same on ARM would require buying a full architecture license (at a price somewhere between $1-10mn) before even starting to design any chip. On x86, it would simply be impossible, meaning third parties have no control over the capabilities of the chip. 

Several key advantages

RISC-V is a Reduced Instruction Set Computer architecture (RISC) like ARM, while x86 is a Complex Instruction Set Computer architecture (CISC). This does not mean that there are fewer instructions or that the chip is less capable: RISC architectures are just based on simpler instructions, thus requiring more steps to compute a complex task but with excellent efficiency, therefore consuming less power than CISC architectures for a given task. This explains the popularity of RISC architectures in mobile applications, where devices are limited by the capacity of batteries. Apple recently brilliantly demonstrated with its M1 family that ARM chips could beat x86 even in terms of performance. Similarly, RISC-V is also simple to program, thanks to its academic origins, and does not have to support several decades of retro compatibility like its competitors thanks to its recent clean-sheet design.

Strong ongoing momentum

The problem traditionally encountered with open-source software is a lack of central planning. In the case of RISC-V, an eponym foundation has been set up in 2015 to standardize the main branches of the architecture, aiming to provide an advanced common technological bedrock and limit incompatibility risks. Among those branches, vector instructions, which are crucial to compute matrices in AI applications, have been standardized in 2021 and will replace custom ones. This makes the platform particularly appealing. As a consequence, the number of memberships is rising fast (+130% in 2021 alone), now totaling nearly 2’500 members and including prestigious names such as Google, Intel and Huawei, a clear sign of the extremely strong momentum of the architecture.

The future is already here  

The price for neutrality is going up

The ultimately failed Nvidia's bid for ARM acted as a wake-up call for many in the industry. The acquisition would indeed have allowed Nvidia to orientate ARM’s roadmap towards its own needs, or even restrict access to parts or to the whole of the technology, something the U.S. behemoth was ready to pay $40bn for. The incorporation of U.S.-based technology also means that it can be subject to political sanctions, especially in the context of rising geopolitical tensions.

While the architecture has generated much interest among private players, RISC-V’s biggest supporter is therefore China, which fears further restrictions to its technologies following the Huawei ban. The country has therefore started investing heavily in the architecture, with the government-backed Chinese Academy of Science announcing it would be releasing a new generation of performance designs every 6 months, and major players such as Alibaba and Huawei announcing significant development efforts.

Intel as an unexpected godfather

Quite surprisingly for the historical champion of the x86 architecture, Intel has displayed a lot of interest of late for RISC-V. The company joined the RISC-V Foundation in February 2022, and announced a $1bn Innovation Fund mainly targeting RISC-V applications. The goal here is to act as a catalyst for the RISC-V ecosystem, notably by supporting design startups, by licensing differentiated RISC-V intellectual property and by providing new generations of modular building blocks (known as chiplets) based on the ISA. Such a support materializes an appreciable shortcut towards establishing a competitive ecosystem and creates a solid bedrock for an efficient dedicated supply chain.

Intel’s announcement is of course not charitable: the company has lost the war against ARM in mobile and is now being attacked in high-performance chips. Subsidizing RISC-V is a conscious strategy to trap ARM between the high-performance x86 and the low-power RISC-V. In addition, boarding early on this train may help become the reference manufacturing player for the nascent ecosystem, keeping its fabs (semiconductor manufacturing plant) humming, generating appreciable revenues, and helping accomplish its recent strategic pivot towards external customers.

Chips already shipping at a frantic pace

RISC-V is today mostly encountered in the microcontroller segment, where it is gaining increasing popularity. Western Digital switched to this architecture for microcontrollers used in DRAM and SSDs, and gave the blueprints to the open-source community. It witnesses strong traction already in automotive, as highlighted by the partnership between semiconductor specialist Renesas and rising star RISC-V chip-designer SiFive to develop high-end automotive applications. The latter has stated that it was within reach in terms of performance of ARM’s A78 chip (2020’s high-end core), and has already validated a design on TSMC’s 5nm process, which will guarantee optimal performances. Other players are developing promising high-performance chips, such as Esperanto (AI accelerators as big as an Nvidia GPU, but consuming 1/100^th of the power) and Ventana Micro (datacenter processors).

The race is on

The race has started for winning a market that may still be small for now (<1% market share), but is growing extremely fast with a 2020-25e CAGR of 115%. This growth is laying the foundation of what increasingly appears as a third way in the computing universe and paves the way for a tectonic shift in the industry. ARM appears poised to be the main loser, while Intel has a major opportunity ahead. In the meantime, RISC-V technical capabilities, low-power focus, and adaptability will mean increasing penetration of this technology in people's daily lives, which will ultimately benefit all our themes.

Catalysts

• Chip flagship product. Commercial availability of a high-end competitive chip in a mass-produced device.

• Software support. Major Operating System (OS) announcing commercial support of the architecture.

• Regulatory or competitive restrictions. Political, regulatory, or business restrictions on x86 or ARM would boost RISC-V's adoption curve.

Risks

• Well-entrenched ecosystems. x86 and ARM have a fully-developed hardware and software ecosystem, meaning there is a risk of strong inertia from users.

• Lack of maturity. Both the RISC-V architecture and ecosystem are only at the beginning of their development, hence potential reliability problems.

• Fragmentation. Due to its open source nature, and despite the guidance of the central Foundation, there is a risk of lack of technological focus leading to ecosystem fragmentation and slower adoption.