The Game Of Ecosystems Intensifies

You may know about my fascination with ecosystems if you have followed my writing. It is only fitting that I am writing this Blog in Munich (shiver, it’s cold), where I attended the GSA McKinsey workshop on “Distributed E/E Architectures and Zonal Computing.” This workshop had attendees from semiconductor foundries, EDA vendors, IP vendors, Tier 1 Semis, Tier 2 Integrators, software vendors, and automotive OEMs. Earlier this week, we announced our extension into the processor ecosystem with our work with SiFive on RISC-V on Edge AI, just months after we had announced our partnership with Arm on automotive applications. As you can tell, we are neutral and work with all processor vendors equally.

So what does the ecosystem look like from the perspective of a System IP provider?

The System IP ecosystem is similar to EDA vendors, and the dynamics are just as fascinating. The following illustration captures the ecosystem’s hardware side:

The System IP Hardware Ecosystem

In the upper half of the illustration, fabless semiconductor companies, integrated design manufacturers (IDMs) that both design and manufacture, and system houses that have taken chip design in-house develop the Systems on Chips (SoCs) that power electronics across the major industries. To get there, developers combine their custom blocks that contain unique differentiation in hardware with reused components on the left side of the illustration. These are design IP, e.g., peripheral interfaces and accelerators for graphics, digital signal processing and AI/ML, and processor IP that handles the computing with all the software in the systems.

Drawing the analogy to the human body, these are the brain, heart, other organs, and interfaces like eyes, ears, skin, etc. On the right side, you find the equivalent to the human circular and nervous systems, better known in SoC design as Networks-on-Chips that carry data (blood) and control (firing nerves). With the help of the tools provided by EDA vendors and service providers, both indicated in the center of the illustration, users take the collection of SoC building blocks, optimize the architecture, verify their integration, and implement them to be ready for manufacturing by foundries shown at the bottom.

Across this ecosystem, the individual providers must agree on, abide by, and partner on connecting using specific protocols and standard interfaces. In addition, physical awareness of IP during the front-end design is critical.

So far, so good. What about the dynamics in the ecosystems?

As indicated at the beginning of this blog, I have been fascinated by ecosystems for quite some time. Developers seek choices, and as already mentioned above, we at Arteris take a distinct neutral position. It does not stop us from preparing the popcorn and observing how processor architecture adoption, for instance, develops over time.

Back in 2014, I wrote in “Game of Ecosystems” about how the battle between different processor ecosystems, including x86, ARM, MIPS, and PowerPC, is like a modern world “Game of Thrones” for the Internet of Things (IoT). Investments in the IoT were aligning around the three flanks of sensor-hub-cloud, with major players such as Intel, ARM, IBM, and MIPS approaching the throne in different ways. At the time, Intel expanded from servers into mobile, while ARM grew the sensor side of microcontrollers and entered the server side with its 64-bit architecture. IBM had a good standing on the sensor side and was quite creative with an open-source hardware strategy for servers called OpenPower, and MIPS held a solid position on the sensor side.

Fast forward two years, in late 2016, I wrote in “Top 7 Verification Trends For 2017—Changes In The Game Of Ecosystems” how significant shifts in the mobile, server and IoT spaces happened in 2016. The mobile domain appeared stable in Arm’s hand, but there were now non-Intel-based servers from OpenPower and ARM. Intel was aligning around servers and IoT, and power, MIPS, ARM, and Intel architectures split the IoT space among them. At the time, I wrote, “RISC-V is also trying to disrupt this space due to its Open Source nature, making it one of the most interesting areas to watch in 2017 and beyond.”

In “Hot Technologies In Cold Weather,” written early 2019, I identified the clustering of ecosystems around processor architectures as a key trend at Embedded World, with the RISC-V ecosystem drawing particular attention.

Today, just four years later, IBM has moved the power instruction set architecture (ISA) via OpenPower into the Linux organization. The Arm architecture has made significant strides in data center computing. We have seen important RISC-V announcements for the data center space during the recent RISC-V summit in December 2022, making it a three-way race of ISAs for data centers. But of course, more and more computing happens at various edges, as outlined in “Hyperconnectivity, Hyperscale Computing, And Moving Edges.” 2014’s simple three-way split into “sensor-hub-cloud” has become much more complicated and opened to more ISAs, more “workload-specific” computing, and configurable processor architectures.

What’s next for ISAs and processor architectures?

There is no clear winner for processor architectures. The game of ecosystems surely will intensify even further. Issues like supply chain concerns during the last few years are toppling design chains in all major areas as system houses re-thinking how to control their destinies – a critical discussion topic at today’s GSA McKinsey workshop, with a fair share of the automotive ecosystem in the room.

And from an NoC perspective? We connect all processor architectures and peripherals and are neutral to our customers’ processor, foundry, and design IP choice. We strive to work with all of them. The announcements we made around the Arm architecture on automotive and RISC-V / SiFive on Edge AI show that we are preparing the pieces to fit before they touch our mutual customers’ development environments. And there is a lot more to come!