Contrary to the current trend, the future CPU will become bigger and bigger. Of course it will also have more transistors, but so the transistor density will also be much lower than today. Why is that? Because they use light.
The idea of a completely lighted computer – or a hybrid of electricity and light is not new, but the latest advances in technology show that a new era with many optically processed functions is come close.
Ability to handle light speed
There are two properties that make optical computers attractive. The first is that it will be especially fast – with the speed of light movement. And when you turn off the light – a feature comparable to that of a transistor in an optical chip – it happens very quickly (think of the femtosecond number, equal to only 1 / 1,000,000 nanoseconds). Combining these two properties can create optical computers much faster than electronic computers.
Its negative aspects are also directly related to the positive side. Using light to turn on and off light is often ineffective, meaning you spend more energy on calculations. Moreover, light moves fast but is also easily scattered around, meaning that the parts must also divide into larger distances.
The middle part is a hybrid device. Light carries information, but switching on and off is done by electricity. Basically, light must be absorbed to create an electric current. After that, the generated current will be used to modulate another optical signal to create an optical transistor.
Materials capable of absorbing light (and producing electric current) are usually quite large in size, leaving room for a large capacitor. The feedback speed of the current will be limited by the charging and discharging speed of the capacitor. This will repeat when modifying the light flow: a block of material must continuously charge and discharge electricity.
Not only does the capacitor charge and discharge consume time, it also consumes energy. While a silicon-based transistor can use about one Joule femto (10-15 J) energy per bit, an optical system can use thousands of times more energy than that number.
That is why optical connections are only meaningful to computers in the data center (or in large systems). But when high performance prevails against energy efficiency, optical connections will make sense for the size of the motherboard. Even so, it is an absolute limit.
Optical diode replaces the capacitor
However, the latest advances of Nanophotonics Center researchers from NTT Corporation, Japan, with the lead of Kengo Nozaki, have solved the aforementioned bottleneck for capacitors and help carry speed the light for these new chips.
Researchers of the project used photonic crystal (photon crystal) technology. In this case, a photon crystal is basically a thin silicon wafer with lots of holes drilled inside it.
Light passing through this silicon blade will stab into the upper holes and be scattered. But the space and the size of the holes mean that no matter which direction the light wave comes from, it will encounter a similar wave that is out of phase with it. As a result, light will be annihilated. In other words, this hole-filled silicon blade is like a perfect mirror.
Illustrative images for optical diodes and images under a microscope.
If a single line of holes is removed from the silicon blade, the light will be directed in the direction of the missing holes. The researchers placed a small piece of light-absorbing material at the end of the waveguide. When light reaches the material that absorbs light, it creates a lot of electrons. This will turn the waveguide into a high-speed photodiode (PD) (the diode is the type of device that only sends current through it in a certain direction).
Researchers have found a way to increase the transmission rate to 40 Gb / sec, which is the same standard for a high-capacity multi-wavelength path. It is worth noting here that this is done with a single wavelength, and this speed is achieved only by a small capacitance of the light-absorbing material.
However, researchers can do more than just increase the amount of data transferred. They also created a device with their super-fast optical diode. To do that, they also put a flexible material with holes completely surrounding it, to create a secondary laser.
EOM (Electro Optical Modulator), converted from an electrical signal to an optical signal.
When powered, the laser emitted from this flexible material will leak light into the second waveguide. Conversely, when a flexible area of a laser has no electrons in it, it acts as an absorbent and absorbs light to the wave path.
The flexible area is electrically connected to the optical diode (connected on the chip, no wires needed). When the optical diode absorbs light, it sends electrons to the flexible area, where it amplifies any signal into the second waveguide. When no light goes to the optical diode, the light will be absorbed in the second waveguide.
Operation diagram of optical transistors. (With PD as optical diode, EOM is an optical modulator).
This is similar to how silicon transistors operate, when electrical signals can be used to control the output and on-off optical signals, as well as amplify that signal to a higher level. Combined with this optical diode and modulator, it will act like an optical transistor.
The researchers showed that they could modulate a signal at a rate of 10 Gbps, equivalent to the standard speed of optical communication devices. They also recognized the capacitance of the optical diode and the modulator below 2fF (femto Faraday), smaller than anything else so far. The smaller capacitance also helps the processing speed faster. However, it seems that optimizing the load current will allow a significant increase in the data rate.
Moreover, it also has a high level of energy efficiency. The researchers showed that their technology consumes less than 0.1 fJ / bit (femto Joule per bit), even smaller than silicon chips.
This does not mean you will soon see optical chips as above – the next step of this will be a hybrid chip. The researchers suggest that their optical transistors will be very useful in maintaining the coherence between caching in multi-core CPUs.
That also means that future chips will be bigger. In the not too distant future, it is likely that certain functions will be shifted to computer parts based on these optical transistors. But certainly not all functions – if an Intel Core i7 processor (with 1.9 billion transistors) is converted into optics, the chip will have an area of up to 48 m2. The balance between speed, energy and size will need to be carefully considered when combining optics and electronics.