The AI can rise strongly in the new century thanks to the breakthroughs in computers, namely industries such as microchips and data storage. But the further the AI industry goes, the more "faster, stronger, and more" the need will be; We need an AI that is both fast and energy efficient.
And to keep up with the speed of AI's run, hardware companies have to show themselves. AI pushes the limits of semiconductor technology, and semiconductor technology must have a new breakthrough to "shoulder" AI, this is a mutually beneficial relationship.
And there is Luminous Computing, the company that has just raised an impressive amount of funds – more than $ 9 million from reputable individuals like Bill Gates or Uber's CEO, Dara Khosrowshahi. Luminous Computing is researching and developing a new chip, while semiconductors often use electrons to transmit signals, Luminous uses light.
In the new era, many industry applications add AI to increase product performance, including automatic cars and unmanned aircraft. However, existing chips are not really suitable for developing AI: they both consume energy, data processing speed is sometimes not fast enough.
These limits will cause the transmission to be interrupted, or "lag" in the language we use. Stalling intervals will make the experience of using AI analyzing data not smooth, and can cause a self-driving car to crash when it is not timely to handle a flash situation.
This "bottleneck" phenomenon is moving in a bad way: new research by OpenAI shows that the amount of energy needed to "train" an AI is doubled every three and a half months. It is unclear whether we can call this a form of Moore's law applicable to artificial intelligence research.
CEO and co-founder Luminous, Marcus Gomez said: The modern world expects a lot of AI, but the limitations of hardware are making the road much slower to develop.
"For years, Silicon Valley has promised us about owning a supercomputer just like a science fiction movie", Gomez CEO said,"but we still have to wait".
A new AI chip that can increase machine learning's power many times, can help us discover new remedies, can use energy so efficiently that smartphones can also run AI. advanced.
Luminous found a way to "lighten" more barking
Researchers at Luminous Computing use lasers as light guides, crept through the microstructure's microstructure. By using different colored light to bring different pieces of data through the chip, the new system possesses superior data transfer capabilities, far superior to the current chip.
The ability to transfer a huge amount of information, quickly shows that optical chips are the leading candidates in AI research. Another "bright spot": they consume less energy.
Mitchell Nahmias, co-founder of Luminous and chief technology officer of the company, said that the current prototype optical chip uses three times more energy efficient than some of the most advanced AI chips at the moment. . All of today's achievements are a multi-year research effort by Nahmias and colleagues at Princeton University.
Luminous is not the only one looking towards light. Other startups like Lightelligence or Lightmatter – the names of both startups contain the word "Light – light" – are also developing optical chips. The chip giant, Intel, is also burning his own optical chip intentions.
Dirk Englund, an MIT professor of technology and lightmatter technology advisor, thinks Luminous will have difficulty adjusting devices to regulate light, especially when entering a mass production process.
Optical chips need both laser and electric-optical modulators to succeed, which is also an obstacle for optical chips to not replace electric chips.
Breakthroughs of AI industry
Billionaire Bill Gates and many other investors are confident that Luminous will overcome the challenge. But their eyes are not only about Gomez, Nahmias and colleagues: will any company penetrate through the narrow gap of the blocked neck, they will reach the true potential of artificial intelligence.
You can imagine, the virtual assistant perspective – Google Assistant, Siri or Alexa – can answer every question you ask for it just a chip in the palm of your hand.
Refer to MIT Technology Review