One major challenge in further developing computing technology is how to overcome performance limitations of CMOS technology. Seeking a solution to this challenge, NTT is engaged in research in the field of photonics-electronics convergence information processing technology, which combines photonics technology transmission with CMOS in order to enable high-speed signal processing that takes advantage of the properties of light. NTT has succeeded in integrating photoelectric conversion elements with the smallest capacitance in the world, using a nanostructure technology called photonic crystals. In April 2019, NTT announced that it had developed an optical modulator and optical transistor that operate with the lowest energy consumption in the world. There are great expectations that if circuits based on this technology can be developed, then there will be potential to realize a never before seen high-speed computing platform with ultra-low energy consumption. We spoke with Dr. Akihiko Shinya and Dr. Kengo Nozaki of NTT Basic Research Laboratories to get an outline of this technology and the concept of the nanophotonic accelerator they aim achieve.
Computing platforms based on CMOS (Complementary Metal Oxide Semiconductors) electronic circuit technology have made it possible to process a large amount of information by improving circuit performance according to “Moore's Law”. However, due to limitations on microfabrication and integrated circuit density, processing with electronic circuits has neared its limits in terms of speed and energy consumption. Although parallel processing with CPUs has increased the amount of computation possible, in recent years, performance has peaked in terms of response delay, which has become a bottleneck in information processing. In the future, there will be higher demand in the world for online processing with low delay and high speed event processing, for use in fields such as security, autonomous driving, disaster prediction, and traffic control. So the question arises of how to solve this problem.
It is expected that the introduction of photonics technology will overcome the performance limitations of CMOS technology. We believe that the solution will be to realize a new computing platform that can perform “photonics-electronics convergence” processing, which uses light not only for signal transmission as has been done so far, but also for signal processing in processor chips linked with electronic circuits. In order to do this, it is essential that we solve the problems of energy saving and chip miniaturization, which have been obstacles to conventional photonic-electric conversion. Our team is trying to solve this problem with low-delay, energy-saving, high-speed signal processing using a nanostructure technology known as photonic crystals that NTT has researched for many years. Here I would like to introduce the “ultra-low energy photoelectric conversion / phototransistor” parts and the “nanophotonic accelerator concept” we created by realizing those parts.
Various research has been done on photonics-electronics convergence, which combines optical technology transmission and electronic technology processing, is one potential solution to the problems in future information processing. In particular, we are aiming to realize nanophotonic accelerators, a type of photonics-electronics convergence processing chip made with nanophotonics technology. The recent realization of phototransistors gave us the idea to propose the concept of this accelerator. We will explain the three elements required for that.
● Optical passgate logic circuit
First is the logic circuit. Light is great for sending information at high speeds over long distances, such as with optical fiber communications. However, information processing has evolved with electronic circuit technology as its standard. Even when trying to use light, the high energy consumption of optical-electrical conversion devices acts as a bottleneck, so it has been considered difficult to do information processing with light and electronic circuits closely linked. In the 1990s electronic circuits started to be made using optical elements, but research gradually declined because these were inferior to CMOS in terms of integration and size. Therefore, we examined whether there is a logic circuit more suitable for light and decided to adopt the optical passgate logic circuit as our method. With CMOS technology, when information (electrical signals) comes in from the top, that information is sent to output by switching logic gates, but when that switching time builds up it causes delays. In optical passgate logic, when information enters from the top, the optical switches are switched at the same time, and the information is transmitted as light to output. We accelerated processing by delegating calculations to the optical circuit that would have been delayed in CMOS. Optical passgate logic circuits can output results with very low delay relative to the input. This principle has the potential to perform not only digital logical operations but also machine learning such as neural networks with low delay.
● Fabrication of nano photoreceptor (O-E) and nano optical modulator (E-O) conversion elements
Low-delay optical passgate circuits are still inadequate for practical use. We will realize low-delay processing by combining CMOS electronic circuits. In order to connect CMOS circuits and optical circuits at high density, we must use an optical modulator that performs electrical-optical conversion (E-O conversion) and a photoreceptor that performs optical-electrical conversion (O-E conversion). We are using nanoscale technology with photonic crystals to make these components far smaller and more energy efficient than conventional technology.
Nano photoreceptors convert light signals into electrical signals. When we embed a functional material that absorbs light into the photonic crystal, optical signals input are converted into electrical signals. On the other hand, if we use a nanoresonator made of photonic crystal instead, it is possible to build a nano optical modulator that converts electrical signals into optical signals. With photonic crystals, it is possible to create a “light insulator” that confines the light to a very small area. This makes it possible to build nano photoreceptors and nano optical modulators with element size and power consumption 1/100th that of conventional technology. We believe that these photoelectric conversion elements will act as an interface to connect CMOS and optical circuits at high density.
Optical transistors serve the role of relaying optical signals by converting optical input signals into a different type of light, then amplifying and outputting the signals. Nano photoreceptors (O-E conversion) and nano optical modulators (E-O conversion) using photonic crystals offer both low power consumption and small size, so they can be closely integrated and make it possible to manufacture O-E-O type phototransistors. The principle of operation for this device is that the nano photoreceptor converts optical signals into electrical signals, and by applying this signal directly to the nano optical modulator, it transfers and outputs the light on a different wavelength. Having verified this OEO conversion operation, we found that the optical control energy used by the photoreceptor was 1/100th that of conventional technology, and the output gain was up to 2.3 times that of the optical signal input. Using the numerical value of this amplification, we were able to prove it was a phototransistor.
These photoelectric conversion elements and phototransistors operate with the lowest energy consumption in the world. Our results suggest that in the future, it will be possible to develop optical signal processing by integrating these parts at high density, including optical passgate circuits. We believe that we can realize a nanophotonic accelerator by combining these three elements and assembling the right peripheral technology on the side of the electric circuit.
Photonic crystals are indispensable for realizing nano photoreceptors and nano optical modulators, and they have a nanostructure with a periodically changing refractive index, so they are well suited for making miniscule structures that manipulate light. Nanofabrication of semiconductors using nanofabrication technology makes it possible to create “light insulators” that confine light at specific wavelengths. NTT has researched nanophotonics for 20 years, and about 10 years ago, progress was made in research on optical switches that use light to control light. It was found that power consumption could be reduced by using photonic crystals, revealing potential for use as a material for industrial technology. Furthermore, in the last 4 to 5 years, progress has been made in research on photonics-electronics convergence using photonic crystals, and it was found that extremely small scale photoelectronic integration is possible, further advancing our research.
Even outside NTT, all manner of research is being done on how to send information inside chips using light signals. However, when it comes to photonic crystals, NTT has a long history with optical communication devices and has accumulated a wealth of processing and manufacturing knowledge, putting us one step ahead and giving us a major advantage over our competitors. Because of this photonic crystal technology it is possible to significantly reduce the power consumption and size of photoelectric conversion devices and phototransistors.
With our newly developed phototransistor, there is the potential for circuit creation, so we presented the concept of a nanophotonic accelerator. At the present stage, research using light to transmit signals inside chips has made progress, and although we have succeeded in making some parts, we have yet to make an accelerator that integrates them. However, thanks to the high performance of photoelectric conversion devices and phototransistors, which have been considered difficult due to power consumption issues so far, it is now possible to use them in circuits. The realization of this phototransistor alone has been a huge achievement, and photonic crystal technology has made it even smaller and more energy efficient. Moving forward, we will enter the stage of working to see how well we can make a circuit based on these three elements and the results of our research so far. It is yet to be determined how many elements can be integrated including an optical pass gate circuit and an optical transistor for meaningful optical signal processing, or how to realize photonics-electronics convergence information processing with CMOS technology. If research progresses and the technology is established, then it will be possible worldwide to focus more on information processing using light. We must make more progress in the electronic development of chips with electronic circuit manufacturers, so we expect that it will take more than 10 years to reach practical use.
As research progresses and light becomes easily usable as an element, the way that element is used will change. If we enter an era in which power-saving, compact light elements can be scattered across the network and easily connected, then it will be the start of a world different from our world where information processing is still done with current computing technology. In other words, as this research progresses, our whole world may change dramatically. We would like to continue our research, so that we can enter a world that goes beyond existing technology.
It seems that it may still take some time for the nanophotonic accelerator concept to be put into practical use, as researchers deal with matters including circuit construction and how to scale up. However, we had an exciting discussion about what kind of world would emerge if it were possible to process information using light in a compact and energy-saving manner, a problem that has been considered most difficult until now. Dr. Shinya told us, “The whole world may change.” So it is no wonder that when the news about optical modulators and optical transistors was announced, people in the industry said “This may be the answer to what comes after the end of Moore’s Law” I look forward to what the world of information processing will be like in 10 or 20 years.
Interview by Kanako Kaisho
on February 25, 2020