Achronix’s Speedster7t FPGA Launches to the Next Level Thanks to Synopsys’ IP
Designing on silicon is expensive, which is why it's important to get it right on your first pass. Achieving just that is Achronix's Speedster7t FPGA for AI applications, with Synopsys' help.
New components for artificial intelligence (AI) applications are popping up more and more often lately. When it comes to AI, your options for computing resources are mainly limited to either GPUs, ASICs, or FPGAs.
One company working in the FPGA camp is Achronix, a California-based company that focuses on delivering FPGA solutions for AI, ML, networking, and data center applications.
Block diagram of the Speedster7t FPGA. Image used courtesy of Achronix
This week, Achronix announced that it achieved first-pass silicon success for its Speedster7t FPGA. By and large, Achronix attributes much of its success to design tools from Synopsys, specifically Synopsys DesignWare IP, which it says allowed it to accelerate time to market by up to 3 months.
This article will look at the importance of first-pass silicon success and the role of FPGAs in the field of AI.
There are no two ways around it: designing silicon is expensive.
Between engineering costs, software costs, and the costs of manufacturing the physical chip, expenses can add up very quickly. The only thing that would make the process more expensive is if the chip didn’t work as expected when everything is said and done.
For this reason, all companies ultimately strive for first-pass silicon success, basically, having their silicon work on the first run. This method requires, amongst many other things, a competent EDA software suite.
Achronix had the fortune of achieving first-pass silicon success on its Speedster7t FPGA and essentially has Synopsys’ tools to thank.
Ways DesignWare IP claims to help with AI design. Image used courtesy of Synopsys
Specifically, the company claims that they were able to reach the market three months faster thanks to Synopsys design tools, citing things like the availability of rich logic library cells and a broad portfolio of memory compilers.
Benefits of FPGAs for AI
Achronix's FPGAs' success could be an excellent thing for the AI industry since there are many reasons why an FPGA is a good option for AI & ML workloads.
The main benefit of FPGAs is that they are both unique (capable of being designed for specific applications) and flexible (capable of being iterated on).
FPGAs are software-defined hardware, where a designer can configure the device's internal logic to be optimized for whatever specific application they need. This flexibility gives an FPGA a considerable advantage over GPUs for AI, where GPUs are more general purpose and were never really designed for AI in the first place.
A general breakdown of FPGAs vs. ASICs. Image from AnySilicon
Since the designer can configure an FPGA in software, it is extremely flexible, easily updated, and iterated, which is particularly important in a field like AI, where "state-of-the-art" is changing every day. All in all, the hardware needs to keep up with the software's ever-evolving needs. An FPGA does this beautifully, while an ASIC, at least the specific chip, cannot be changed once it's made.
Along those same lines, FPGAs can help designers bring products quicker to market versus the standard ASIC design process, which, again, is invaluable in a dynamic and ever-changing field like AI.
By achieving first-pass silicon success thanks to Synopsys' design tools, Achronix can further push that state of AI hardware. Its success could be critical since FPGAs are uniquely positioned to economically enable AI solutions and create new additions to the marketplace, which are invaluable.
It would be interesting to see Achronix's FPGAs in various AI applications, hopefully breathing life into new, more scalable hardware solutions.
Featured image used courtesy of Achronix