To the Editor

Caulfield and Dolev's recent Nature Photonics commentary1 provides an interesting perspective on the potential role that optics may play in the field of supercomputing. I wish to comment that there are a number of key impediments for the use of optics in digital computing that perhaps demand a much more guarded view of the ability of optics to compete with digital electronics. Caulfield and Dolev point out that optical interconnects will help to reduce communications bottlenecks and argue that optics will also extend into the realm of digital processing. The importance of optical interconnects is widely acknowledged, but the potential of optics in digital processing is much more speculative. I'd like to focus on a few key issues.

First, logic devices and circuits. Any form of scalable digital computing requires nonlinear elements to process digital data. A number of key functionalities are required of the nonlinear elements. These include logic-level restoration, cascadability, fan-out and input–output isolation. Transistor circuits deliver all of these functionalities in electronic computing, but for large scalable logic circuits, no optical element or circuit, either active or passive, can do this and at the same time compete with transistors in the key metrics of energy consumption and small device footprint. Digital optical computing will become competitive only if there is a major breakthrough in the functionality, energy consumption and footprint of nonlinear optical devices. In this context, Miller2 has given a circumspect view of the prospects for the invention of an optical transistor.

Second, fan-out and voltages. Caulfield and Dolev claim that optics has two important advantages over electronics: 'superior fan-in and fan-out, and signal transmission without the need to apply a voltage'. The first claim is difficult to justify, given the fact that today's nanometre-scale CMOS electronics, which pervade large-scale computing, are not limited by fan-in and fan-out.

The second claim is correct, in a strict sense. Propagation of an optical field requires an optical carrier to be applied. But a voltage transient applied to a transmission line also propagates by itself. Importantly, energy is required in both cases. Voltage transients have an advantage over optical fields in that they can be very easily interfaced directly to that most remarkable of devices — the transistor. What is missing in Caulfield and Dolev's claim is an acknowledgement of the simplicity and elegance of baseband digital electronics such as in CMOS circuits. Here, the voltage supply rail is a resource that is called upon only when required; current (and therefore energy) is drawn only when needed.

I suggest that apart from the advantages of low-loss signal transmission, there is an inherent disadvantage in using an optical field as a basis for representing logic levels in computer circuits. Note that optical circuits use one or more carrier waves, with data modulated onto those carriers. This is the same concept as the radiofrequency electronics used in wireless communications. In contrast, digital electronics uses baseband data that extends down to zero frequency in the electromagnetic spectrum. In principle, one could build a digital electronic computer using modulated radiofrequency carriers, just as Caulfield, Dolev and others propose using modulated optical carriers for computing. However, no electronic circuit designer would do this, simply because computing with baseband data has overwhelming advantages.

Third, passive elements and energy. Caulfield and Dolev claim that 'passive elements for manipulating light ... require no energy to be supplied'. Unfortunately, all real-world passive elements, whether optical or electronic, exhibit losses and therefore consume energy. In fact, with the exception of free space, there is no such thing as a zero-energy passive element. To properly evaluate the energy consumption of any optical circuit, it is necessary to take into account the energy consumed by losses in passive devices as well as the energy consumed in the all-important active devices. Incidentally, electronic circuits abound with passive elements such as wires, capacitors, inductors and parasitic elements. Electronic circuit designers must often carefully evaluate losses in these elements because in large circuits even tiny losses can add up to something significant. The situation is exactly the same in optical circuits.

Fourth, speed. Caulfield and Dolev argue that 'the speed of optical processing is ultimately limited by the speed of the electronic input and output'. It is important to recognize that the clock speeds of today's processor chips are limited by energy consumption, and not by the speed limitation of individual transistors. For this reason, the processor industry has suspended the trend to higher frequencies in favour of parallelism. Even if a practical optical transistor is invented at some point in the future, the processing speed of a chip using that transistor may also be ultimately limited by energy. Whether that chip can compete with an electronic chip will depend on a number of factors other than speed, such as its device footprint and energy consumption.

Fifth, memory. The lack of a satisfactory optical memory is a serious impediment to the development of optical computing. Caulfield and Dolev comment that the apparent speed of light can be altered 'in some striking ways'. Unfortunately, slow light waveguides are bulky and consume significant energy3. Microring resonators have similar energy consumption issues and an even larger footprint4. Digital electronic memory is tiny and has vastly superior performance.

Sixth, cost. A major advantage of transistors in today's electronic chips is that they are very small — of the order of tens of nanometres rather than the micrometre or larger scale of optical dielectric waveguides and other optical devices. This has a major bearing on the exceptionally low cost of integrated electronic devices which, in turn, has contributed to their dominance in the world of computing.

In summary, it is important not to lose sight of the astounding advances that continue to drive digital electronics on an ever-improving trajectory, and the very large challenges that face digital optical processing.