Artificial intelligence has made headlines all across the tech world in recent years. Whether defeating humans in Go or Jeopardy, making new advances in speech recognition, or just crafting trippy illustrations a la Google’s Deep Dream, artificial intelligence is changing our world.
A Battle of Hardware
Tech companies have invested heavily in AI in hope of becoming the leader in the field. AI systems consist not only of software, but of highly sophisticated hardware capable of supporting the calculations required to run such a program. It is for this reason that many companies have invested in developing their own chips to support their AI development.
For example, Google’s Tensor Processing Unit (TPU) is an application-specific integrated circuit, or ASIC, designed to handle deep neural nets. A network of software and hardware that learns specific tasks through the analysis of large amounts of data, it is essentially one chip designed to perform one specific function really well. In order to perform a different function, an entirely new chip must be designed.
In the chip-making industry, there is an arms race to become the leader in AI hardware technology. Traditional processors will not be able to fulfil the demands of advanced AI algorithms, and it is up to hardware manufacturers to design new technology that will meet the needs of tomorrow’s intelligent devices.
But the path to this brave new world is not entirely clear. AI is in its infancy and new advances like deep neural networks present essentially uncharted territory. While Google uses its custom-made TPU to handle these new systems, other companies like Microsoft prefer to use processors called field programmable gate arrays, or FPGAs. Still other companies use machines endowed with massive amounts of GPUs, which have been shown to handle the demands of AI much better than traditional CPUs. Everyone is exploring seemingly endless approaches to AI hardware, and they are hungry for new and better ones.
While the chips in question are made of tried and true silicon, the size, shape, density, and placement of these chips will radically change, to optimize for AI calculations. This will require some changes to the ways electrical engineers use near field measurement tools for things like EMI compliance testing.
A Looming Revolution
This represents an impending shakeup in the hardware industry. The tech giants have vast online operations, for which they buy and operate more hardware than anyone else in the world. The continued and increasing importance of cloud computing will only make these networks grow larger, and any choice that one of these tech companies makes regarding their AI hardware will cause a fundamental shift in the chip industry.
For example, because Google’s TPU is manufactured by Google itself, if they continue to use it for their AI operations to the exclusion of third-party chips, this can pose a major threat to companies like NVidia or Intel, two of the leading chip manufacturers. If GPUs prove to be the best way forward for AI hardware, NVidia, the primary manufacturer of GPUs, stands to benefit immensely. By the same token, Intel owns Altera, the company that sells FPGAs to Microsoft, so if those end up dominating the market, Intel will be positioned as the leader in AI hardware.
And all this is to say nothing of future advances that might be made. Who knows what new technologies will revolutionize the industry? New materials like graphene promise to potentially change the fundamental ways in which chips operate. Though advances in machine learning and neural nets have shaped the growth of AI so far, we have still only scratched the surface. And what about the impact of smartphones and other portable devices? Only time will tell.
Artificial intelligence is already here. It’s changing the world around us, and it’s only getting started. From offices and factories to hospitals and homes, AI will revolutionize the way we work and live our lives. For the hardware industry, radical change is coming sooner rather than later. But it will come for the rest of us soon enough. All we can do is pay attention, sit back, and let the chips fall where they may.