If you take a piece of paper and fold it in half, it doubles in thickness from a tenth of a millimeter to two-tenths. Fold that same piece of paper in half over and over another thirty times and it is thick enough to reach from the ground to the International Space Station; a distance two billion times larger just from thirty-one folds. Fold it another twenty-two times and it is poking into the sun. The exponential growth shown in this process is almost unfathomable, yet it is exactly what computers have been doing for the last fifty years. They have sped beyond imagination and have brought about a technological revolution beyond all others. Computers have gotten so much faster, in fact, that current smartphones are more powerful than room-sized supercomputers in the 1980s (“The future of computing”). Many people are familiar with this themselves, even joking that computers are outdated by the time they leave the store. This started about fifty years ago when the prediction that computational speed would double every two years, termed Moore’s law, was first shared with the world. Since then, the prediction held true nearly every year and computational speeds have skyrocketed. Computers are at the forefront of technology; nearly everything we have today either relies on computers or was designed through one. The continuously increasing speed has been so consistent that people have grown to rely on it. What then will happen when what we have come to expect every year is no longer achievable? The answer to this question is both impending and seemingly unavoidable. As the fundamental part of all computers, transistors, grows increasingly smaller, it is becoming physically and economically impossible to create them any smaller. Once they can no longer become smaller, the computer sector will become hard-pressed to continue to make computers faster. Within the next few years, Moore’s law will end from the impossible smallness of transistors and new methods in computing will be explored instead of just shrinking current technologies. Software engineers will have to further optimize code, computing architecture will need tweaking, and computers may have to find new ways to interpret and compute information.

In April 1965, co-founder of Intel Gordon Moore predicted that the number of components that could fit on a standard computer’s central processing unit (CPU) would double every year (Moore, 116). For ten years, his forecast held mostly true, but with more data he changed his original one year doubling time to two years. Even after the modification, this forecast was still particularly audacious yet held mostly true for the next forty years. The forecast and its longevity are impractical to relate to other technologies. If the speed of cars followed this same trend, the fastest car today would go about sixty-seven million miles per hour; another seven years and it would surpass the speed of light (“The future of Computing”). The trend has led the world into technological growth “enabling a technical, economic, and social revolution never before experienced by humanity” (Denning, 54). According to Waldrop, none of the computational speed we have achieved today was inevitable. Every time the chip-making companies would release the next stage of computer technology, software developers would create applications that “strained the capabilities of existing chips” (Waldrop). When the chips were unable to handle the load placed upon them, consumers were prepared to pay for the next-generation chip that could handle what they wanted of it. People relied on the increasing computational speed and because of this reliance, Moore’s law became a self-fulfilling prophecy. People saw it as a roadmap for future chips to follow, and often times companies such as Intel and Advanced Micro Devices did too. They would set up their future plans based on them reaching and potentially surpassing the prediction set forth by Moore. This strategy was termed “More Moore,” named after Moore’s law. They would strive to meet the prediction and most times did. This prediction has brought the transistor count in a CPU from 2300 in 1971 with the Intel 4004 to nearly seven billion in 2016 with the Xeon Broadwell-E5 and this number continues to grow (economist). While slightly behind the 2016 estimation of ten billion, the prediction held mostly true up until recent. The larger the transistor count, the more the next generation has to improve. Continuing the two-year doubling time is becoming more and more of an impossible task.

There are several issues the computer industry is having to go against in improving the processing speed of a computer. The CPU, where these problematic improvements are mostly focused, is almost like the brain of a computer. Most of what a computer is doing at any given time is done through the CPU. It is, in essence, just a large collection of transistors connected in ways that allow for logic to be performed. Modern Intel processors have around seven billion transistors all packed onto a chip about half a square centimeter. The CPU sends an electrical pulse through the system and the transistors use this pulse to do whatever operation it is trying to do. This could be adding a number or determining if a word is spelled as it is in the dictionary, etc. The CPU can send this pulse more than a billion times a second; modern CPUs send an average of up to three to four billion pulses a second. For several years, companies tried to increase clock speed to help make processors faster and this worked for a while, but as the clock speed approached 3.5 Gigahertz (billion times per second), the amount of heat produced became too much for the system to handle (Waldrop). It was no longer feasible to increase the clock speed. To get around this, companies started creating multicore CPUs; essentially packing multiple processors into the one CPU. Two processors going half as fast as one processor could ideally get the same amount of work done as one, and it produced a bit less heat. This, however, became unfeasible as too many processors also started to cause overheating and if the dissipated heat cannot be removed from the system it will damage the CPU. While these techniques were successful in managing issues for a time, they are no longer viable and the industry is running out of alternatives.

The physical issues behind making transistors smaller are only part of the problem. To add exponentially more transistors onto a chip costs a lot both in research and in having to create new fabrication lines every time the transistors get smaller. The photolithography machines that create the processors are expensive and each time the transistor size is halved, a whole new fabrication facility to make processors needs to be designed and created (Waldrop). This typically costs billions, and few companies can afford to do this sparingly. Fortunately, Intel is promising to persist, suggesting that they would have one hundred billion transistors on a processor by 2026 and stating that they are only increasing the doubling time from two years to two and a half years. While not true to Moore’s 1975 prediction, it is still an expectation that the exponential trend will continue despite the increasing issues. This promise may be made more realistic by the advent of new manufacturing techniques. One promising example, extreme ultraviolet (EUV) lithography may allow lithography machines to robustly perform the etching needed for infinitesimal transistors and could allow processor manufacturers to continue without as heavy of a burden from the high costs. The current machines rely on lasers to do the etching, which have a small wavelength, but the components needed to be etched are getting smaller than the wavelength of the laser light etching them (Waldrop). Continuing to use this same size wavelength would be like trying to make an inch-wide hole in a wall by shooting a cannonball at it. To make more small and accurate cuts, EUV lithography relies on ultraviolet light which has a much smaller wavelength. Should this technique work and be utilized, Intel can be expected to follow the trend for just a bit longer, and it is very likely that competitors will try to follow.

Even if companies can manage to get through all the issue presented to them, there is one looming issue that is simply unavoidable: transistors are approaching the fundamental limits of physics. Currently, high-tech transistors are around fourteen nanometers across (Waldrop). This size is already incredibly small, but by the early 2020s it is suggested that transistors will be around three nanometers across. At this point, the transistors are about ten atoms across and are bordering on completely unreliable. Transistors are essentially just electrical gates (SciShow). If an electron is being pushed through a transistor, it can only get completely through if the transistor is “opened.” When the “gate” is only a few atoms in size, however, the electron can perform a phenomenon known as quantum tunneling. The electron can sometimes simply go through the transistor without it being enabled. This is technically possible at any scale, but only at this infinitesimal scale is it anywhere within the realm of a probable occurrence. If even one electron can get through a transistor, whatever program, document, etc. that is being worked on can become completely inoperable. If a transistor is vulnerable to this phenomenon, it simply cannot be used within a computer. There is some hope of different materials such as molybdenum disulfide that somewhat thwart the quantum tunneling silicon transistors experience, possibly allowing transistors to get as small as one nanometer (Yang). These technologies, however, are not guaranteed to work and would only delay the physical limit still quickly being approached. This is simply an unavoidable issue and new methods and/or technologies will have to be made practical if the computer industry has any hope of continuing to improve.

Many people anxiously await the newest gadgets and technologies so they can marvel at the ingenuity of their designs and take advantage of all they have to offer. When getting a new device, people often simply assume that it will be faster than the previous. The public become trained to expect the continual increases in performance and capacity as this as always happened in the past. People often expect the newest devices will simply be faster. The technology economy has relished in the high profits of a society that covets the newest state of the art technology. This craving, however, will leave us starved if we only rely on what has been done in the past. The days for the perseverance of silicon transistors are limited. Fortunately, there are many promising technologies (Markoff). One of the most well-known is likely quantum computing. While not overly economical, quantum computing may allow for a vast increase in computational speed magnitudes over what current technologies can do. An example of this is the D-wave 2000Q. A quantum supercomputer that can do certain computations up to ten thousand times faster than current computers (D-Wave). So far, however, quantum computers are nowhere near convenient enough for the average consumer to use(Waldrop). Another potential technology is neuromorphic computing which tries to model the processing units after the human brain. The brain is able to do about one quintillion operations a second with twenty-five watts of power while modern high-tech processors can do a much less efficient three hundred billion operations per second with one hundred forty watts of power (Furber). This puts the brain at nearly twenty million times more efficient than a modern high-tech processor. The brain is much slower in doing certain things, such as adding/multiplying numbers, but if it can be more closely modeled in computer systems they could potentially be a lot more efficient. Unfortunately, similar to quantum computing, this also has not made it outside of labs much, but the speed of the processor is not the only thing to rely on. Software developers have always relied on the speed of computers to allow for more and more power to be supplied to their programs. They generally would just allow the brute force of processors to allow for quick and easily built programs. Why make neatly written programs when the processor can compensate for quickly done programming? As an example, consider multiplying one thousand times itself. A simple program can simply ask the processor to go through and multiply the numbers. While effective, it could potentially be done more efficiently by simply telling the computer that it can add the zeros up since the number is a factor of ten. Realistically, the program would take no more than a hundredth of a second to do both, but at a larger scale the difference becomes more drastic. Programmers will be required to think through their program and optimize it to use what they have available from the processor instead of allowing the processors’ exceptional speed to handle it. Luckily, many programs that require vast amounts of computational power are able to simply send requests to “server farms” commonly known as the cloud (Waldrop). These server farms can contain vast quantities of processors that can get large amounts of information processed and can simply send finished calculations back to the program. Consequently, mobile phones and computers will not have to continue to get faster for every application to become faster and more productive. As long as transmission speeds can stay fast and reliable, these server farms can be used in most cases. These are not good for everything, however, and have to be carefully maintained and updated. This method works for companies that can afford the costs of hosting a server but other programs whose companies cannot afford this and programs that process too much raw information to transmit reliably will still have to rely on local computer processors. This is only really good for applications that give little data but require a lot of computation with the small amount of data. Examples being an AI such as Siri interpreting voice input or doing large mathematical problems.

Many are optimistic that quantum computing will be the savior of computational speeds. Whereas typical computers work in zeros and ones (bits) for processing data, quantum computers work in zeros, ones, and the superposition of both zero and one. This allows quantum computers to process several different states for each quantum bit as the superposition of the bit can hold more data. Imagine it like using a light to transmit information. If the light can be either on or off like a conventional system, then information can be sent via Morse code. But, if the light can be on or off or be any shade of gray, each shade could represent a letter and information could be sent much more easily and quickly. Because quantum computers can hold more information in a single piece of data than current computing systems, it is able to do the work of current computers several magnitudes faster. This technology, however, is unconventional to say the least. Because of how sporadic quantum information can be, the processor has to operate in near perfect conditions. The most modern quantum processor D-Wave 2000Q is about the size of a thumbnail (D-Wave). But, since the processor has to be in perfect conditions for it to be useable the actual system is about the size of a small car. It has to run at -273 °C to slow the particles down so they can be read. To remove interference, the quantum computer is shielded fifty thousand times stronger than Earth’s magnetic field. On top of that, the processor is held in a nearly perfect vacuum and the system consumes 25kW of power. It is simply too hard to produce quantum computers at a consumer scale that are better than current computers within the next decade. While the modern computers we have today did start in a similar condition, they have had around sixty years to get to where they are now. Quantum computers simply cannot be made reliable quickly enough. 

No matter the case, the life of the silicon transistor is nearing its end. Already, the transistors are so small that companies are dishing out billions to get them smaller, but no matter how much money is put into new assembly lines transistors can only get so small. Moore’s law will likely hold longer than many suggest as new materials are becoming more successful in allowing smaller transistors, but this will only delay the inevitable. Transistors need help. Fortunately, manufacturers can make some more improvements in processors that do not require making the transistors smaller such as improving architecture or investing in stacking technologies for placing CPU cores on top of one another. This is not the only hope, either. Since server farms do not have a size and heat issue more server farms can be implemented (so long as transmitting capabilities continue to improve) meaning less computational speed will be needed locally. While this will not work for everything, it is going to become more prominent. In the coming years, some creativity is going to be necessary in keeping up with the increasing demands. Fortunately, speed is not the only aspect of a computer. A Mac and a Windows computer, for example, may run at the same speed yet both have advantages over the other; they are not the same. Many aspects not reliant upon computational speeds can continue to be improved. Also, as everything is/already has gone mobile, a lot of future work may be centered more around adapting processors to a mobile environment. Often consumers do not care much about the speed of a phone so long as it is adequate for what they need it to do. If the phone is fast enough, a long-lasting battery becomes a more desirable trait. There are already devices whose processing chips can perform tasks with so little power that they can sit indefinitely using only power gathered from outside power sources such as motion and heat. If this level of efficiency can be adapted to processors within mobile devices, those devices could theoretically go magnitudes longer without being charged. Using as little power as possible while still being fast is becoming the next big goal. It is not fully clear what is to come of the processors of the future, so for now software developers will just have to practice optimizing and hope for breakthroughs in computational technologies. 

