Chips in Artificial Intelligence

Chips in Artificial Intelligence


In March 2016, after 4 Go matches, AlphaGo, Google’s artificially intelligent Go-playing computer, beat Lee Sedol, Korean grandmaster with 3:1 and currently it ranks No.2 in worldwide Go Rating.


This historic Go match set a new exclamation point for the discussion of artificial intelligence (AI). Many are surprised at how advanced AI has developed and also how people react to a machine in the most complicated board game in existence.


AI is not new and tech giants are investing heavily into this field, hoping to become the leader in this technology. The AI system not only consists of software, it also requires high hardware functionality to support the calculations. This is why many companies are investing in developing their own chip designed to support AI development.


Google, for example, has designed an ASIC or application-specific integrated circuit that is specific to deep neural nets, an AI technology that is reinventing the internet services. Google named it Tensor Processing Unit, or TPU. It is a network of hardware and software that can learn specific tasks by analyzing giant amounts of data. One drawback of ASIC is that it can only perform one function really well. If you want to change function, you have to redesign another chip. When asked why they are not using FPGA (Field Programmable Gate Array), which has more flexibility in reprogramming, Google says “FPGAs are much less power efficient than ASICs due to their programmable nature. The TPU has an instruction set, so as TensorFlow programs change or new algorithms are developed they can run on the TPU.”


Another company who has invested over $2 billion in research and development of its AI chip is Nvidia. On April 5th, Nvidia released a new chip called Tesla P100 that is designed to provide more power for deep leaning, an AI technology that produced Goolge’s AlphaGo. The chip has more than 150 billion transistors, which makes it the world’s largest chip. A neural network powered by Tesla P100 can learn from data 12 times as fast as it was possible by Nvidia’s previous best chip.


Other tech giants are also investing in AI through acquisition. Twitter, despite its weak stock price, announced its purchase of Magic Pony, a British artificial-intelligence start-up for $150 million. This is the third machine-learning start-up Twitter bought after Madbits and Whetlab. Microsoft bought SwiftKey this Feb, hoping to get into the AI field with predictive search technology brought by SwiftKey.


For chipmakers, the hardware for AI is definitely the field that everyone wants to win. Traditional Intel-based x86 processors will not be able to support AI algorithm as Graphic Process Unit (GPU). This is definitely the trend in the semiconductor industry we should look out for. Soon, we might see more and more powerful, yet energy efficient chips in the market to power various advanced devices.


We, at Advanced MP Technology, keep a close look at the electronic industry for the newest technology and trend. By doing so, we are able to adopt our services to match with the technology development to serve our customers.