Intel faced a big hurdle recently when they reached the tipping point of the number of transistors being added per square inch to their processor chips. It seem that they have finally hit the the limit of Moores Law, A Law founded by Intel cofounder Gordon Moore in 1965, who noticed that tranisistors were shrinking so fast that every year the number of tranisitors you could fit onto a square inch of a chip doubled. This trend has continued to this day, although has recently slowed to every two years. As you can expect this law cannot continue at this rate forever (some transistors are smaller than a virus) and it seems the restraints of the physical world have began to kick in. But fear not, amazingly Intel has recently made a massive breakthrough with how to combat this limit of nature that Chrisopher Roland has stated "could be one of the most important of our generation,".
While Intel keeps going forward with its CPU technology each year, performance has continued to increase by a fair percentage, a feat in itself, but the actual thing of wonder is this new revolutionary breakthrough. What exactly would stir up the CPU segment and would actually set the benchmark for the future of computing? We will take a look at Silicon Photonics, our topic of interest here.
Before we go dissecting this huge topic, I reckon it is fair for us to do a little revision on what semi-conductors actually are and a little overview on basic definitions associated with this topic.
What Are Semiconductors?
To give a simple definition semiconductors are a substance that can conduct electricity under some conditions, but not others, hence they are a great medium for controlling the flow of electricity. In terms of a physics definition, Semiconductors are those elements who have a minimum energy gap between their valence band and conduction band. An actual conductor has zero energy difference between these bands which gives free electrons for the flow of electric current. While an insulator (something that doesn't conduct electricity) has a very large gap.
In semiconductors there is a small energy gap, not as great as that of insulators or zero like in conductors. A minimal amount of energy is required for the Valence Electrons to jump to the Conduction Band and cause the flow of current. Silicon in this case is a well known semiconductor alongside Germanium and the other elements of Group 4 on the Periodic Table.
There are two types of semiconductors. First is Intrinsic, it is a semiconductor made of only one element. Then there is Extrinsic, this semiconductor consists of impurity elements from Group 3 and Group 5 elements. Basically Group 3 elements are trivalent, basically having a "hole" where an electron can enter from a element having an excess electron, which in this case is the Group 5 element. It is referred to as Pentavalent.
So Where Is The Limitation?
In essence there are 2 limitations, one lies with the material being used as the semiconductor which is Silicon in this case. And then there are the connectors of the semiconductor which can also be referred to as interconnectors. They basically supply electricity and are a bridge for data transfer in and out of the different semiconductor components of a system. This is where copper wires come in.
I'll be elaborating on the problems of the latter case because this is the prime area which Silicon Photonics will address and in short: be revolutionary.
You see Copper wires act as pathways for the flow of data as well as electricity to different integrated circuits and components of a system. But not only is copper "inefficient" but it also puts a limit on the thinness of the wire, after which you'll have to use a different wire for faster flow. This creates a big bundle of wires regardless of them being extremely small, increasing the amount of power consumed and the amount of heat dissipated by the entire system.
And this Ladies and Gentlemen is the reason why all supercomputers require a huge amount of power and expensive cooling solutions. Tianhe-2, which is a 33.86-petaflop supercomputer (fastest supercomputer on the planet to be precise) requires around 17MW of power to run it and an additional 9MW power for its cooling.
There is also the limitation of Moore's Law because believe it or not, a point will come when a transistor will be equal to the size of an atom. And cutting an atom in half would only result in a nuclear explosion, something you don't really want to do in a computer. This theory was presented by Dr. Bernie Meyerson, IBM Research Fellow.
And Now Silicon Photonics
Silicon Photonics is the method of transfer of data using optical rays. Just like optical fibre networks are rapidly replacing copper installations throughout the world in the field of telecommunication, we can expect the same advantage in computing.
Initially the concept of Silicon Photonics was deployed to large data centres and places where you need insanely fast speed for huge amount of information to be transferred. Traditional Copper wire impose a huge limit on the speed, meaning nowadays we rely heavily on optical fibers for our home ISPs (Internet service provider). At higher speeds, you get a smaller and smaller Sound to Noise Ratio on copper wires unless you restrict their size and use optical fibers for long distances.
While this discussion is getting sidetracked into telecommunications, the point is that Silicon Photonics can be used to increase the efficiency of a modern computer to a completely new height. Just like I said before, the losses due to heat and noise can be avoided plus we would be able to make computer components even smaller in size. Nowadays around 11 pico joules of energy is used to transport 1 Bit of data (11 pJ/Bit), it is estimated that when IBM and Intel will fully roll out their solutions to the market, say by around 2020+, the energy consumed will be massively reduced to around 250 femto joules per bit (250 fJ/Bit) which is a staggering 0.25 pico joules.
With that explained I would like to highlight the top 2 companies which are working on this concept right now. Intel with designation of Silicon Photonics for data centres and IBM with the designation of Silicon NanoPhotonics. Both have promised insane speeds going beyond 100GB/s.
Let's discuss the Intel side exclusively now, Intel researchers have predicted the future of computing lies with this concept. One of their stockholders actually described it as a "chip with frickin' lasers". And you know what, he has pretty much summed up this entire article in four simple words.
The idea is to create a macro-chip. Intel is looking to give it's Xeon CPUs a boost. They are looking to integrate the silicon photonics concept using an Altera FPGA to control the high speed interconnects going in and out of the chip.
Currently they are limited in processor manufacturing by two main factors: Power and number of cores per chip. By using the low power but high speed interconnect, they could bundle in more chips, separate the L1 cache from the CPU, connect it using the interconnect and then use the available space for more components. You could disaggregate GPUs, minor CPUs and other components in the same manner.
Their stockholder predicts that this new technology could potentially give AMD a run for its money if it is implemented on the server end of the market. I believe we should all expect Silicon Photonics based processors to hide the end PC users shortly after that too.
So GD'ers that was a fairly long read I think. Now I want to hear from you, what are your thoughts on Silicon Photonics? Can it be a potential replacement of Moore's Law? Sign off in the comments folks!
- Intel’s ‘Miraculous,’ ‘Mind-Blowing’ New Chips, Per Susquehanna on Barron's
- Will silicon photonics replace copper cabling in mainstream datacentres? on ComputerWeekly
- IBM announces silicon photonics breakthrough, set to break 100Gb/s barrier on Extreme Tech