The Growth of FPGA as AI Hardware: Adaptive Computing

Source : DRex, Chosun Biz.

The versatile semiconductor known as Field-Programmable Gate Array (FPGA) is gaining attention, particularly in fields requiring high-level computations such as Artificial Intelligence (AI), Big Data, Cloud, Autonomous Driving, and High-Performance Computing (HPC). While Intel and AMD dominate over 90% of the FPGA market, both companies have achieved record-breaking performances in the past quarter due to the increasing demand for FPGA.

According to the semiconductor industry on the 12th, FPGA stands out from Application-Specific Integrated Circuits (ASICs) like Central Processing Units (CPU) or Graphics Processing Units (GPU) since it allows for the internal circuits to be changed based on the intended purpose. This adaptability is beneficial in AI applications, where technological advancements occur rapidly, as FPGA can alter the circuit structure through software updates without requiring a complete redesign as with general-purpose processors.

In contrast to GPUs, which are not originally designed for computations and are deemed inefficient in terms of cost and power consumption, FPGA demonstrates fewer shortcomings in these aspects. Particularly, FPGA can be customized to suit AI models, leading to enhanced power efficiency. It allows not only parallel computations but also high-performance basic operations, making it suitable for both "training" and "inference" in deep learning applications.

The FPGA market is expected to grow at a CAGR (Compound Annual Growth Rate) of 14.2% driven by the growth of the AI market, reaching 15.5 billion dollars by 2027.