Plugging the current leak: Moore's Law continues to work in transistor development

In 1965, Gordon Moore predicted that the number of transistors that can fit on a chip of a given size would double every two years. This is known as Moore's Law. This law has been working for many years. The first integrated circuit (invented by Jack Kilby of Texas Instruments, see picture) was a clumsy big thing, and now transistors are measured in nanometers (one billionth of a meter). People have created fast and intelligent computers with beautiful patterns and connected the world at the speed of Moore's Law. Since Dr. Moore established this law, mankind has entered an incredible era of information technology. What a seemingly inadvertent discovery has such a strong vitality.

It is not really a law, but a phenomenon, a description of the long march of technological development, each step of which involves specific technological changes (see chart). The unstoppable momentum of technological development has become a prophetic creed. Every "shrinkage" of transistors is a step towards their minimum size. If this law continues to develop, within 20 years, transistors will be the size of a few single-crystal silicon atoms.

To be more precise, transistors have become so small that every atom in a space of this size becomes crucial. Too few atoms and the insulation between them disappears, or current leaks to places it shouldn't flow due to "quantum tunneling" (a phenomenon in which electrons disappear and reappear somewhere else). Too many atoms of the wrong kind are equally bad, affecting the conductivity of the transistor. So engineers are working hard to redesign the transistor. It seems that Moore's prediction will continue to be valid for some time to come.

Atomic nucleus motherboard

A transistor is actually an electronically controlled converter that consists of four parts: a source (where current flows in), a drain (where current flows out), a channel (connecting the source and drain), and a gate (which controls the switching of the channel through changes in voltage ) . In traditional transistors, these components are all distributed on the same plane. To prevent leakage, one idea is to change the transistor to a three-dimensional design.

Making a transistor stick out of a mother chip allows for a more efficient arrangement of many of its constituent atoms, especially those that make up the channel and gate. Extending the channel outward and surrounding it on three sides with gate atoms increases the surface area of ​​the gate, allowing for better control of the channel and reduced leakage. The better the function of the transistor's gate in the on state, the more current it can pass.

In May, Intel, the famous American chip giant (Dr. Moore is also a co-founder of the company) announced a plan to commercially develop the technology, which is marketed as "Tri-Gate". The company expects to launch the new transistor later this year. This transistor uses half the power of existing transistors and is particularly suitable for use in laptops. After all, battery life is a major selling point of laptops.

It is difficult to promote the full use of three-dimensional models in a mature industry, after all, their two-dimensional models are already mature. The Silicon on Insulator Consortium, which includes the US company Globalfoundries and the British company ARM, is trying to improve the flat transistor as one of their alternatives. The consortium's technology is to make transistors inside a thin layer of pure silicon wafer. Under this layer of pure silicon is an insulating layer, and under that is a standard wafer. This standard wafer is used as a substrate to place transistors. This method requires the channel of the transistor to be made thin enough so that the electromagnetic field generated by the gate can pass through the entire channel, increasing the maximum control that the gate can exert. But this method forces the Silicon on Insulator Consortium to face the second problem caused by the continuous reduction in transistor size: there are either too many or too few atoms that deviate from their normal positions.

To improve electronic performance, the silicon used to make transistors is often doped with other elements. The latest transistors are so small that doping their channels requires only a small amount of impurity atoms injected into the silicon. If this amount is not well controlled, the normal operation of the transistor will be affected. But deviations in the manufacturing process make this requirement difficult to achieve. The process of doping the ultra-thin channel that the Silicon Insulator Association wants to use is extremely difficult, so they decided not to dope the silicon with impurities, but to make the transistor channel out of pure silicon. But this requires that the silicon layer cannot be thicker than 5 nanometers. And this thickness must be almost the same across the entire wafer. Intel (admittedly, it is not a calm bystander) believes that such a precise standard will inevitably increase the manufacturing cost of transistors.

SuVolta, a small company in Silicon Valley, has proposed another approach. The planar transistor channels they plan to manufacture also do not contain impurities. However, the company intends to use inexpensive conventional silicon wafers without changing the composition of the wafers or making the ultra-thin channels required by the Silicon on Insulator Association. Their unique feature is to add a gate underneath the channel. The two gates work together to control the channel, which has no impurities added and is not thin enough. In this way, a transistor with better functions and lower energy consumption is created. The company says that its energy consumption is reduced to only half of that of conventional types of transistors, without any loss in performance. SuVolta's move has aroused great interest from Japanese electronics giant Fujitsu, which now has a production license for this technology.

How much room for development is there?

All of these approaches mean that Moore's Law will continue to work for at least a few more years. Hundreds of experts update the International Technology Roadmap for Semiconductors every year . They predict that the lateral dimensions of standard transistors will shrink to 16 nanometers by 2013 (it's 32 nanometers today) and to 11 nanometers by 2015. Going even smaller will require a conceptual leap. Fortunately, there are already several options for that.

One of the most promising approaches was outlined last year by a group at Ireland's Tyndall National Institute, led by Collinge, who published a paper announcing that they had created a contactless transistor. The method was patented in 1925 by a physicist named Julius Lilienfeld, but until now it has been difficult to make.

The two sides of the transistor junction are silicon doped with conductive electrons (because electrons carry a negative charge, it is called n-type material), and the p-type region is doped with positively charged holes in the lattice, which are generated by the ionization of electrons. There are also some transistors where the source and drain are both p-type and the channel is n-type. In other cases, the situation is just the opposite. At the junction of n-type and p-type, the silicon acts like a valve to prevent current from flowing in the opposite direction.

However, as transistors get smaller, the more difficult it is to make the PN junction, which is also affected by fluctuations in the concentration of the doped elements. Dr. Kolinge's design - similar to Intel's tri-gate, which wraps a three-dimensional gate around a single, ultra-thin silicon wire - avoids this by making the entire transistor out of a semiconductor that is doped with a higher concentration of elements than the semiconductor used in conventional flat-panel transistors. The design contains an extremely thin channel that acts like a valve, in which all carriers (for example, free electrons or holes) disappear when the circuit is off and are filled with such carriers when the circuit is on. Its size should also be able to be reduced. Researchers at the Tyndall Institute reported last year that computer simulations of this atomically arranged contactless transistor showed that they worked perfectly and had a gate length of only 3.1 nanometers.

This gate length will allow Moore's Law to continue to work in the next few years. After that, Moore's Law will require more innovative thinking to continue to work. For example, a large number of academics and engineers are thinking about how to make a transistor that makes the quantum channel a feature rather than a defect. According to quantum theory, electrons can only be obtained at a certain energy level, which means that transistors using the tunneling effect may directly switch from weak current to strong current without warm-up time.

It may be a good idea. In a world where the size of transistors is limited by the size of single atoms, it is unclear whether this is the last improvisation of engineers. When Dr. Moore announced this law, he thought it might be valid for 10 years. The irresistible force of human ingenuity has ensured that Moore's Law has lasted much longer than expected, but this force is now facing the insurmountable obstacles of atomic physics. It is a fascinating race.