The emerging Internet of Things (IoT) industry is an aggregation of products and services, complementing each other to enable efficiency and cost-optimization in multiple industries. It does not have a vertically oriented value chain. It stretches horizontally across multiple industries and markets such as Industrial automation, Automotive, Medical, environmental monitoring, and many more. The use-case for these industries are quite diverse. At the front-end of the applications are end-nodes or sensors, which monitor the ambient conditions and pass on the data down the chain. These end-nodes will be scattered in billions in various industries.
Rise of Custom processors
As mentioned in my earlier post ARM vs Intel: The new war frontiers, COTS processors will not be ideal for building these end-nodes, as the latter are application-specific. Companies would be inclined to custom-processor, as it offers flexibility to assemble only required parts. These parts can include analogue sensor, DSP or proprietary IP, etc. Further, custom processors substantially reduce BoM cost and die-size, which will minimize power dissipation. It also helps companies to differentiate their product from those of competitors. The low entry cost and pervasive presence of IoT industry will encourage many start-ups and small companies to build products for niche applications. With custom processors, these companies can further optimize the cost.
Failing Moore’s Law
In addition to the proliferation of the IoT devices that will give a huge boost to the custom processors, another influencing factor is questionable existence of Moore’s Law. For more than five decades, Moore’s Law acted as a self-fulfilling prophecy. Semiconductor companies raced to make this law a truth, irrespective of whether the market needs high performance processors. There are always the innovators and early-adopters that are desperate to use products based on leading process nodes. However, the mass market takes time to switch to these new products. With high-performance, reduced power consumption, and cost reduction, Moore’s Law ensured that technology played a dominant role.
However, currently the economic equation that is guaranteed by this law, is failing. Leading process nodes are becoming complex to design, with long lead time for commercialization, thus the cost equation does not hold true. The quest for cost optimization forced the industry to look for alternatives, as shrinking nodes is no longer economically beneficial. Customization is the answer, as it can reduce the BoM cost significantly. The billions of end-nodes does not need leading process nodes, custom processors at matured nodes will be good enough.
ARM is a monopoly in smartphone processor market. In the embedded and IoT space, there is no dominant architecture yet. ARM is ideally poised to fill up this vacant position, as it already has a strong presence with CPU IP offerings at diverse power, performance, and price options. With DesignStart license for Cortex-M0, ARM has enabled custom processor design at low cost with less risk. This program will be really useful for start-ups and small companies, as they get an access to proven architecture and IPs at low licensing cost, complemented along with an extensive ecosystem of IPs, software support and silicon partners that can massively reduce the time-to-market for products.
How we can further optimize cost of custom processors?
Open Source Software (OSS) has played a vital role in democratization of the software industry. One of the most popular examples of OSS is Linux operating system. OSS enables innovation and differentiation, at a low cost of adoption. This promotes small firms, start-ups to build products based on OSS such as Linux. A large community of developers supports the software development, so there is no risk of vendor-locking or obsolescence of proprietary technology. The collective efforts of the community ensure a large ecosystem, and this benefits all the users. Linux has gained enormous presence in diverse applications such as embedded, PC, etc. Network effects can also be leveraged, as more users start using Linux, more features, and utilities are added.
RISC-V extends the open-source movement into CPU ISA. It is an open-source ISA that is license-free and royalty-free. As RISC-V is void of any licensing, the ISA can be used for building custom processors with zero licensing cost. RISC-V is gradually building an ecosystem. In Embedded World 2017, RISC-V showcased the extensive ecosystem with FPGA solutions, security IPs, debug infrastructure, etc.
Few ARM customers have already started using RISC-V for designing custom processors. Now, SoC design companies can develop custom processors at a low cost, without the paying licensing rents. With some NRE investment, these firms can develop the SoC and get it manufactured in fabs. Thus, the price of the processor will be lesser than those based on ARM IPs. At the face value, it looks like we have found an ideal candidate that has the potential to become the dominant ISA for IoT industry. With customization & zero licensing cost, RISC-V looks like a winner.
Is it really that simple?
Are there free lunches?
Linux is a quite successful with billions of deployment in diverse products. Although, it needs considerable effort and expertize in using Linux for commercial products, the benefits weigh over the man hours. Linux offers unmatched flexibility. The enormous community provides a good ecosystem around the OS, with extensive support for peripherals, 3rd party software, etc.
However, extending the concept of open-source to chip design is a different ball game, owing to the basic differences between software and hardware. Unlike software, which needs only time and effort to be developed, hardware involves tangible components, for which someone has to pay. Secondly, you can rework on the software any number of times after testing on the hardware, emulator, etc. Bugs can be fixed with only investment in time and effort, maybe with minimal cost. However, bugs in the hardware can be a million dollar loss! Multiple spins of a processor can shoot cost substantially. Finally, hardware design is more complex than software development.
Let us consider the case of open source RISC-V. In a SoC, CPU IP is just one part; there are many other physical IPs and peripherals needed. So, an extensive IP and EDA ecosystem is needed around the CPU IP. You can only get the CPU IP without licensing rents; however, the surrounding ecosystem is gone cost you. IP vendors should see a viable business case to add support for RISC-V in their portfolio. Let us assume that a strong community backs the RISC-V, and it offers all the IPs and tools needed for building SoCs. However, the question remains whether companies building custom SoCs, will take the risk of using a community backed ISA? Bugs can lead to multiple tape-outs, which add huge cost. Finally, designing a SoC is complex and needs good expertize in multiple areas such as implementation, physical design, packaging, etc.
With ARM ISA, most of the issues mentioned above are alleviated. You get access to proven IPs, robust ecosystem (software, cloud services, security solutions, silicon vendors, fabs), and committed support, instead of community support offered by open-source ISA. Design complexity is reduced; however, still some expertize in SoC design is needed for building custom processors.
Who will build RISC-V based SoCs?
The idea of open-source is disruptive, as it enables a level-playing field to companies, with limited budget, to compete against big players. Although the concept of open-source ISA is revolutionary, it may not have a disruptive effect in democratizing chip design.
In my view, it is unlikely that small companies and start-ups addressing some niche application in the IoT space, will invest time, effort and money in building custom processors based on a community-backed ISA, as they have to validate whether the entire system meets their specifications. It would be a safe bet for them to use licensed ISA, as they get a proven system, complemented with a robust ecosystem. Multiple tape-outs of the SoC can add substantial cost. A matured ISA with some initial cost would be a good starting point, instead of a free fledging ISA. SoC design is not their core activity, so hiring a diverse team for chip design, may not be a pragmatic decision. ARM is used widely across the industry, so the design part can be outsourced to some small companies, specializing in ARM-based SoC design. EDA tools and fabs are costly. EDA vendors and fabs already support ARM-based IPs; they should see economic benefits for adding support for RISC-V. Until RISC-V reaches a critical mass adoption, it’s like a chicken-and-egg situation. Multi-homing adds cost for any company be it a fab, EDA vendor, design firm, or an app developer. Low volume business will attract higher rents. All these cost overheads have to be factored while building RISC-V based SoCs.
Market leaders in the SoC design will definitely develop RISC-V-based SoCs, as it increases their buying power by having some alternative to ARM. However, I believe that these companies will not be interested in engaging with low-volume customers, who needs custom processors. Owing to their large overhead, it makes business sense to sell millions of standardized SoCs.
Summing from above, in my view RISC-V, in its current state, cannot significantly disrupt the semiconductor market structure. One of the key virtues of open-source movement is minimize entry barriers into a market by offering a good enough base, in comparison to licensed entities. Although, RISC-V will offer flexibility for building custom SoCs at low cost, the ecosystem is not yet ready to accept it. The whole semiconductor industry need to work in sync to make RISC-V successful.
It is too early to pronounce the verdict on RISC-V. Any new idea takes time to flourish. I am sure the people at RISC-V are smart enough that they would have foreseen the issues that I have mentioned above and many of the issues would have been already mitigated internally. In my view, RISC-V should focus on one segment such as IoT end-nodes or something else, and then offer this segment a compelling and complete solution, along with a holistic ecosystem, rather than focusing on the entire IoT and embedded industry. Once they achieve mass adoption in one segment, it is easier to spread into other segments, as new users have a great case study or example to look upon.
What should ARM do better to be perceived as a leader in the embedded and IoT segments? I do not have any answer for this, as from an external perspective things look pretty well now for ARM, with a huge installed base. The upward trend will continue in future as well. Extending DesignStart license to other Cortex-M IPs would be a good option for further adoption. However, the main forte should be the strong ecosystem of OS support, cloud services, security, IPs, debug toolchain, EDA, silicon partners, etc. All these play a vital role in building a successful product based on a custom processor at low cost.
Low-cost and customization are often mutually exclusive. Any ISA addressing both these ends will play a dominant role in the IoT industry. My views are limited to my knowledge. I believe there will be many lacunae in this post, so I look forward to the improvement areas. I do not have any professional obligations toward any companies mentioned in this post. The views expressed are completely personal.