2020 will be the year the silicon industry takes notice of some quiet trends and realizes they will completely reshape the industry. It will become apparent that the so-called “end of Moore’s Law” does not define a new pace for the industry, but rather serves as a wake-up call that opens our eyes to other scaling opportunities.
Architecture: If you’ve heard John Hennessy or David Patterson talk recently, you’ve heard about the “new golden age for computer architecture.” With transistor scaling failing to keep us on the curve, we are forced to turn to architectures that offer greater levels of parallelism than CPUs — GPUs, TPUs, and others we don’t even know about yet. We must look to domain-specific and application-specific architectures to solve our greatest challenges.
FPGAs: While ASICs are a worthwhile investment for traditional architectures, for rapidly-evolving application-specific architectures, reconfigurable silicon, such as FPGAs, can offer reasonable tradeoffs. Architectural benefits can reach the market faster with less expense at lower volumes and can be iterated often on the latest fabrication technology.
3D: Fabrication benefits include 2.5D and 3D scaling. We tend to think of Moore’s Law in a two-dimensional sense, but the third dimension gives us a whole new scaling factor as we incrementally conquer various challenges.
Cloud: Cloud computing is transforming the world we live in, and it is impacting the silicon industry in several major ways.
-
The focus on security will continue.
-
Cloud-based EDA is finally beginning to gain some traction.
-
Many of today’s ASICs are built for massive data center deployments. This means a single customer can determine the fate of an ASIC project, which affects job security and disrupts organizations. To keep resilient, chip companies are utilizing more contingent workers, which is problematic in teams with deep “tribal knowledge.”
-
Cloud providers care deeply about optimizing their hardware for their own needs. With increasingly deep pockets, they have begun to build their own design teams and produce their own silicon, giving chipmakers reason to rethink their place in the world.
The impact of the cloud will escalate in 2020.
Cloud FPGAs: The cloud and FPGAs are colliding in a way that will shake this industry to its core. FPGAs have been available in the cloud for a few years now, bundled with commercial EDA tools, on a pay-per-use basis. After some initial excitement, they’ve made little more than a splash because a new ecosystem takes time to evolve. But the opportunities are substantial. Web and cloud applications have reshaped the software industry, and now cloud FPGAs enable hardware developers to participate in that transformation. FPGAs, implementing application-specific architectures, can be integrated as accelerators for web and cloud applications for computationally-intensive tasks in situations where internet latency and bandwidth are not bottlenecks. GPUs and TPUs can be similarly integrated. Combined with browser support for WebAssembly, the performance situation for web technology versus stand-alone applications will completely reverse.
Open Source Silicon and Startups: But the most transformational surprise resulting from cloud FPGAs will be the rise of open-source silicon and a fresh startup culture. We’ve seen RISC-V take off in unprecedented fashion due to the mere fact that it is free of proprietary entanglements, but there is much more to come. While open-source developers have completely transformed the software industry, the community of silicon hobbyists is much smaller due to the expense and complexity of silicon and EDA software. Having FPGAs and EDA software available in the cloud significantly reduces the expense barriers and standardizes the platform. FPGA development now looks almost like software development and the silicon industry will very quickly play catch-up with the software industry, becoming nimble, collaborative, and prolific. In time, open-source FPGA development will lead to open-source ASIC development, which is already getting a hefty boost from government funding. Hobbyists, academics, and startups will find new opportunities to drive innovation that will threaten the stronghold of industry giants.
Design Methodology: The importance of developing a wide variety of new and application-specific architectures, as well as the new involvement from hobbyists, puts substantial pressure on an already urgent need for easier and more productive design tools and methodologies. After 70,000x of Moore’s Law scaling, our 35-year-old RTL design methodology is beyond its breaking point. SystemC, OpenCL, and HLS have appeal for folks entering hardware from a software background, and the urgency for abstraction will lead to further adoption. But many hardware engineers will continue to resist C++-based abstractions that obscure the implementation. A few commercial teams have started fresh with academic languages like Chisel. Redwood EDA is all-in on TL-Verilog to address this need with less disruption. As these technologies jockey for position, EDA and silicon will see new life, and the era of RTL will come to a close. Design teams, at least those that will survive, will benefit from a new level of productivity.
Open-source EDA: Significant EDA advancement is coming recently in the form of open-source solutions, rather than proprietary ones. Open-source EDA is at an inflection point. FPGA designs have recently been implemented using 100% open-source tools, and the US government is providing significant funding for open-source EDA. While physical tools are not yet competitive with industry tools from a QoR standpoint, front-end tools will see increasing commercial adoption. Just as software developers expect compilers to be free and open, hardware developers will adopt similar expectations.
2020 will bring a foreshadowing of industry transformations to come. Cloud FPGAs and open-source EDA will open doors for a burgeoning open-source silicon community to develop open IP and for new startups to develop application-specific architectures to address the next level of computational challenges. Chipmaking giants will face unfamiliar pressure from startups on one side and absorption by cloud providers on the other. Survival in our future industry will require an understanding of web and cloud technologies, a familiarity with the open-source ecosystem, and a willingness to change. One thing that will not happen in 2020 — this industry will not slow down!
