Search

ASICs, Artificial Intelligence and Neural Networks

Updated: Feb 12


Progress in AI and Neural Networks are growing at breakneck speed due to the huge amounts of investment and research being dedicated to the segments. Neural networks are the core of modern AI, where models are trained and applied to specific tasks. A good example would be the automotive industry, where major manufacturers are working on building deep neural networks for autonomous driving.


As neural networks get deeper and more complex, the data processing requirements are also stacking up, with data processing across cloud, supercomputers and enterprise markets being forecast to grow into a $50 billion market. As neural network research evolves, computing power requirements will continue to get exponentially higher as larger data sets are used and the amount of layers increases.


Autonomous Driving


As mentioned earlier, autonomous driving systems are a major use case that companies are focusing on. Developers are constantly progressing towards higher levels of automation. In the autonomous driving world, there are 5 levels of automation: Level 1 refers to simple driver assistance technology while level 5 is complete automation where no human attention is required, even to brake or steer. As developers work towards the goal of level 5, they must overcome 2 major hurdles. The first is that they must be able to process huge amounts of data from multiple sources locally and in real-time. The second is that they must cater to the power-hungry nature of GPUs and CPUs currently being used for this processing.


This Brings us to ASICs


ASICs are microchips designed for a special application. They can be specifically designed to meet the needs of automotive systems and can not only be both powerful and extremely energy efficient but also allow an automotive client to differentiate itself from the rest of the pack. They provide design flexibility and allow for designs that are much more powerful than current GPUs.


Tesla is a good case-study, where they dropped Nvidia’s GPU system and developed its own ASIC chip. Elon Musk pointed out that the transfer between GPU and CPU ended up being a key constraint, with their ASIC performing more than 10x faster.


While ASICs are the ultimate end product, FPGAs are the tools used for prototyping or validating a concept. It is not recommended to prototype design with ASICs since once the silicon has been taped out, usually nothing can be done to fix a design bug.


Pipelining


There are many steps involved in the creation of an ASIC for a specific purpose. From planning the processing pipeline and charting out the hardware module layout to fabricating the chip. Many tests, adjustments and licensing come into play throughout the process as well and these steps directly translate into a long duration (ranging from months to years) and financial commitments as large minimum order quantities are normal with the best fabrication foundries such as TSMC and Samsung.


Throughout this process, software development and system integration cannot proceed.


FPGA based prototyping


A customised prototyping system, made with multiple FPGA modules solves the above-mentioned problem. Companies like proFPGA design ASIC prototyping systems to allow for early software development and real-time system verification. Their systems are highly customisable and modular to match a potential ASIC end-product and allow development to proceed.


With the use of such a system, product manufacturers will be able to capitalise on what would usually be downtime while waiting for the ASIC end-product to be ready. Being able to concurrently work on the software significantly improves the allocation of resources as well as shortens the product time-to-market.


proFPGA has recently released its next-generation prototyping system based on the recently released Intel® Stratix® 10 GX 10M FPGA, which is currently the biggest FPGA on the market. The new system handles complex ASIC and SoC Designs for up to 2 Billion ASIC gates and gives design and verification engineers unprecedented speed and flexibility for high-speed verification and bug hunting to shorten the time to market by eliminating costly re-spins and by providing early prototypes for software development and/or to end customers.


Dana Labs is proud to be the official distributor of proFPGA systems for Southeast Asia. For more detailed information or a casual chat, please drop us a note.

© 2020 by Dana Labs Pte. Ltd.