Artificial intelligence (AI) semiconductors will have explosive growth as their application expands to the edge from data centers, a Samsung executive said.
A new AI called GPT-3 was recently launched in the US, Changkyu Choi, executive vice president at Samsung Advanced Institute of Technology, said at a local industry forum. The AI uses 800 semiconductors.
Using the cheapest cloud available still costs US$3 million, Choi noted.
A cheaper AI semiconductor with high performance will allow to offset these costs, the EVP said.
The relatively expensive GPU were used for AI tasks but new chips such as neural processing unit are being launched that are designed specifically for AI applications.
More system-on-a-chip on smartphones have NPU embedded in them, Choi noted, which means more smartphones can use AI without going through the cloud.
Sensors on smartphones will need to get smaller while their inference capabilities must also increase, the EVP said.
The high power consumption from AI use is also another obstacle that must be resolved, he said.