Computer processors are an integral part of the modern world. Most, if not all of our day-to-day lives require processors to run smoothly. On board computers in modern cars, phones, tablets, shop tills, smart TVs, other smart Home technologies and so on, all require processors to function.
The constant push for faster, more efficient technologies helps to drive research and development in the area of semiconductor and microchip technologies that will continue to allow technology to improve our lives for years to come.
Processors in a computer are made of different parts that perform different functions. Each part is a chip made of integrated circuits built from semiconductor wafers that are connected with copper or aluminium wiring. A wafer is just a thin slice of semiconductor material. Semiconductors are materials that allow conduction of electricity more easily that insulators e.g. Wood, but not as easily as conductors such as copper. The semiconductors and wiring are used to produce components that make up transistors and this allow different computing processes to occur.
The metal oxide semiconductor field effect transistor is the most produced artifact in human history, with production since it’s creation exceeding MOSFETs. It was invented by an Egyptian engineer at Bell Labs in the 1950s.
In 1965, Intel co-founder Gordon Moore estimated that the number of transistors that could be fit into an integrated circuit would double every year, and in 1975 revised this to claim that the number would double every two years. However, this has become a sort of self-fulfilling prophecy as Moore’s Law as it was dubbed, is still used for long term R&D planning.
Around 2010, reports began to suggest that advancement in semiconductors was beginning to slow down, yet if you glance at a plot containing the transistor count over time for processors, there does not really seem to be a significant shift in the rate of progress.
Moore himself expects that his Law will cease to apply around 2025.
The Huawei Mate 40 and the Apple iPhone 12 were released in late 2020, sporting chips with transistors that were only 5nm across. It is expected that the first commercially available products containing 3nm chips will be available in 2024/25 , with the next expected size being 2nm. However, experts do not know whether transistors smaller than 2nm will be possible (Consider that 1nm is roughly the width of only five silicon atoms). Currently, the transistor components are so small that any defects (which are a common thing in materials) such as missing atoms can hinder the transistor’s performance.
Other potential issues include the density of wiring in the wafers. Now that there are millions of transistors in each square millimetre, there are vast amounts of wiring required in these chips, and as the transistor count increases, so too will the amount of wiring, and in spite of increasing clock frequency within the chips, there will be a slowing of transit of the signals produced by the chips due to the amount of increasingly narrower wire the signals need to transverse for an operation to occur.
As the size of processors continues to decrease, and the density of transistors in them continues to increase, the energy consumption of the chips are not decreasing proportionally. This is Dennard Scaling, a hypothesis that ran parallel to Moore’s Law that stopped describing the trend in energy consumption of these components in the 2000s. As a result, in order to maintain performance and efficiency, some transistors in these chips must be left unused. This effect means that the CPU is slowed down and as a result, IBM and Microsoft believe that multi-core processors will be no more due to this effect.
Physical constraints are also at the mercy of economic constraints, as some potential solutions could present more inconvenience for the consumer than is worth it, or would require increasing chip size, which would then defeat the purpose of decreasing chip size for cost and energy efficiency. However decreasing wire cross section will increase the resistance in the wires, so even spacing out smaller wires would present problems.
For the time being, it seems that we must continue with our dedication to the status quo and continue in the pursuit of miniaturisation.
This then begs the question, what will happen when we reach the atomic scale and cannot make transistors smaller? Many large businesses rely on the ‘outdating’ of old models to continuously generate profit. When we reach the limit of transistor size, this will no longer be the case, and will likely force many businesses to either adapt and restructure, or develop new ways of maintaining continued growth and improvement.
One consideration would be to be rid of transistors and use something that more suits our purpose. However, thus far, the study of solid-state devices has been incapable of producing a device that could succeed the transistor.
Making technology larger to facilitate greater large chips and hence greater computing power also has it’s limits. Phones and laptops are popular for their convenience and so making them larger defeats their purpose. As a result, soon we will need to turn to New methods in computing
We must somehow adapt the transistor or replace it if we wish to maintain the continued rate of advancement in technology. In saying this, using the lasers that we use, we expected to run into issues when the transistors decreased below 65nm in size, however New methods of allowing us to use the lasers accurately has meant that this unsurpassable barrier has been long forgotten about.
Regardless of whether we were capable of overcoming the previous issues we came up against, What ideas do we have that could present a possible solution?
In 2016, professors at Stanford university and the University of Texas successfully produced a transistor gate that was only 1nm in size. This gate was produced from a transistor that was made from molybdenum disulfate () and carbon nanotubes as opposed to the standard silicon. The reason that this is possible as the has a lower di-electric constant (roughly 4) than Silicon (roughly 11.7) the study performed showed promising results, however scalability and ability for mass production was left in question but the study, as certain properties of the material were prone to cause unwanted effects.
Another proposed theory has been that reading an electrons spin state provides information in the same way that on/off gates in transistors do. Electrons have a certain property known as spin, which is measured to be either “spin up” or “spin down”, and using this could hypothetically be a solution to our enigma. However, again, some issues have arisen, those mainly being that transistors are extremely useful since they at capable of changing resistance in the circuit by switching from one state to the next. However, when this would be attempted by a system in which electrons are used as the bits, the difference in resistance in different states produced by the components that measure the spins is minimal.
Further alternatives include quantum computing. With quantum computing, like with the electron spin measurement alternative, we deal with spin states. However, for this proposal, instead of having a binary system of states, the information is stored in bits that are a superposition of states, that is a combination of spin up or spin down states, and so the parameters that describe the system are continuous. This means that the computer has much greater capacity than classical computers. However, this also has limits as simple systems are described by vast amounts of parameters. If you wish to build a system that has some practical purpose, the number of parameters required becomes huge, meaning that quantum computing is impractical with current technology.
Yet another alternative proposed has been to use light in CPUs in order to vastly speed up systems and decrease energy expenditure.
Optical components are difficult to make as they require what is known as quasi-particles in order to mediate the signal. In most cases the set up is required to e maintained at extremely low temperatures, however IBM’s most recent offering has succeeded in operating at room temperature. Within this device, switching times of below a pico-second were achieved.
Many discussions about the viability of these devices have focused on the high energy requirements of optical logic. It is possible that the technology can e used in order to mitigate the predicted drastic increase in energy loss arising from increasing lengths of chip interconnects.
In another paper a team of researchers successfully produced an optical transistor using a single dye molecule. The set up required cryogenic cooling however, and also cannot be scaled down below the size of a micron.
Using Cadmium Sulfide, researchers from Penn university[4] have successfully produced a photonic transistor with similar properties to those of the electrical transistors used in devices nowadays. The authors of that particular paper believed that using the devices that they had developed that on-chip photonic devices were very much a possibility in the near future.
The scaling down of transistors may possibly e continued through the use of caron nanotubes. These are tubes produced from graphene and are very conductive. They have a much higher carrier mobility than silicon FETs and so could e more energy efficient and faster, however they are not without their complications.
For example, there is a layer of insulation between the gate and channel known as the gate dielectric. The issues arising in silicon transistors due to tunnelling effects were dealt with using hafnium dioxide, however in Caron nanotube transistors creating dielectrics thin enough to control the devices proved difficult. In the end the researchers were ale to produce a dielectric only 4nm across that had an on off ratio similar to that of silicon devices.
One other issue with this is that the nanotubes are difficult to reliably orient and pack densely, however researchers in Peking University had promising results regarding an experiment where they successfully packed 200 nanotubes per micrometre in alignment in 2020.
Power dissipation occurs due to high resistance between the nanotubes and the contact metals. Some research has been undertaken to try to eradicate this problem, however minor success has only been achieved with p-type devices. Doping has been shown to decrease the resistance using a process known as molybdenum oxide y a factor greater than .
Small percentages of nanotube transistors are metallic, and the presence of these metallic transistors can lead to high leakage currents and even possibly incorrect logic functionality. This is certainly not wanted in computers and other devices and so more work is required in order to rectify these issues. All in all, Carbon nanotubes present a very interesting case and do have the potential to revolutionise the semiconductor industry, however this will ve some years away yet!
Quantum computing meanwhile uses the concept of superposition to perform computations. When compared to the standard on and off setup of the current transistors, the states in a quantum computer are continuous defined through combinations of different vectors. The result of this is that a very small quantity of qubits would be sufficient in order to surpass the capabilities of a classical computer.
The issues arise from this in that in order for the system to remain isolated from the surroundings it requires superconductors that must remain at roughly 77 Kelvin. Ambient temperature alternatives are being produced but they are even more costly. Qubit interaction is being promoted through the use of entanglement, however it is unlikely that we will see a quantum computer available for consumer use in the near future!
Again economics will always be a limiting factor in the production of these devices and the decreasing prices of high speed processors isn’t helping the case for high-cost quantum computing.
References:
https://arxiv.org/pdf/1408.3821.pdf
https://science.sciencemag.org/content/354/6308/99
http://www.fisgeo.unipg.it/luca.gammaitoni/fisinfo/documenti-fisici/physical-limits-silicon.pdf
[] https://www.nature.com/articles/s41566-019-0392-8
[] https://ee.stanford.edu/~dabm/379.pdf
[7]Hills, G., Lau, C., Wright, A. et al. Modern microprocessor built from complementary carbon nanotube transistors. Nature 572, 595–602 (2019). https://doi.org/10.1038/s41586-019-1493-8
[1]Dürkop T, Getty S A, Cobas E and Fuhrer M S 2004 Nano Lett. 4 35
[2]NSM Archive – Physical Properties of Semiconductors. www.matprop.ru.
[6]Molybdenum oxide on carbon nanotube: Doping stability and correlation with work function, Rebecca Sejung Park et al, Journal of Applied Physics 128, 045111 (2020)
[5]Franklin, A. D.; Chen, Z. Length Scaling of Carbon Nanotube Transistors. Nat.
Nanotechnol. 2010, 5 (12), 858–862.
[4]Aligned, high-density semiconducting carbon nanotube arrays for high-performance electronics, Lijun Liu et al, Science 22 May 2020: Vol. 368, Issue 6493, pp. 850-856 DOI: 10.1126/science.aba5980
Low-Temperature Side Contact to Carbon Nanotube Transistors: Resistance Distributions Down to 10 nm Contact Length, Gregory Pitner et al, January 24, 2019
Nano Lett. 2019, 19, 2, 1083–1089 January 24, 2019 https://doi.org/10.1021/acs.nanolett.8b04370
https://doi.org/10.1038/s41586-020-2441-3
https://doi.org/10.1088/0034-4885/61/2/002
https://arxiv.org/abs/1801.00862
Leave a Reply
Want to join the discussion?Feel free to contribute!