July 17, 1997

Transistors of the future: Will diamonds be an engineer's best friend?

Clive "Max" Maxfield, Intergraph Computer Systems

Semiconductor-fabrication processes are approaching their natural limits. The electronics industry will have to start developing and using new materials and technologies to keep up with the increasing need for smaller, faster transistors.

If there is one truism in electronics, it's that faster is better, and the industry invests a staggering amount of research and development in increasing the speed of electronic devices. Ultimately, there are only two ways to increase the speed of transistor switches based on existing semiconductor technologies. The first is to reduce the size of the structures on the semiconductor, thereby obtaining smaller transistors that are closer together and use less power. The second is to use alternative semiconductor materials that inherently switch faster. For example, the band-gap effects associated with GaAs's 3:5 valance structure mean that these transistors switch approximately eight times faster and use one-tenth the power of their silicon counterparts. However, GaAs is a difficult material to work with, whereas silicon is cheap, readily available, and relatively robust. Additionally, the electronics industry has billions of dollars invested in silicon-based processes and would be reluctant to leap into something outrageously new unless extremely compelling benefits were associated with doing so.

For these reasons, designers have traditionally achieved speed improvements by making transistors smaller. However, we are reaching the end of this route using conventional technologies. At one time, the limiting factors appeared to be simple process limitations: the quality of the resist, the ability to manufacture accurate masks, and the features you could achieve with the wavelength of ultraviolet light.

Around 1990, when structures with dimensions of 1.0 µm first became available, technologists believed that 0.5 µm would be the effective limit they could achieve with optolithographic processes and that the next stage would be a move to X-ray lithography. However, there have been constant improvements in the techniques associated with mask fabrication, optical systems and lenses, servo motors, and positioning systems. Also, re-searchers have made significant ad-vances in chemical engineering, such as chemically amplified resists, in which the application of a relatively small quantity of UV light stimulates the formation of chemicals in the resist, which accelerates the degrading process. This approach reduces the amount of UV light necessary to degrade the resist and allows the creation of finer features with improved accuracy. The combination of all these factors means that 0.25-µm processes started to come online early this year, and it is now feasible to achieve structures as small as 0.1 µm by continuing to refine existing processes.

The speed of a transistor relates to its size, which affects the distance electrons have to travel. Thus, to enable transistors to switch faster, technologists have concentrated on scaling, or reducing the size, of the transistors. However, as you reduce the transistor structures, you must maintain certain levels of dopants to achieve the desired effect by increasing the concentration of dopant atoms. Increasing the concentration beyond a certain level causes leakage, however, resulting in the transistor's being permanently on and therefore useless. (You can't call it a "switch" if it's always on.) Thus, technologists are increasingly considering alternative materials and structures.

Heterojunction transistors

You can use new technologies that center on the type of junction between semiconductor regions to create transistors that are faster than existing ones. An interface between two regions of a semiconductor having the same basic composition but opposing types of doping is referred to as a "homojunction." For example, consider a generic NMOS transistor (Figure 1). Assume that you're dealing with a positive-logic system, in which a logic 1 value has a more positive potential than a logic 0. In this case, when you present a logic 1 value to the transistor's gate terminal, the gate terminal's positive potential (relative to a logic 0) repels the positively charged holes in the p-type material, thereby opening a channel and allowing current to flow between the source and drain terminals. In this type of transistor, you form all the doped regions in the same piece of semiconductor, so the junctions between the n- and p-type regions are homojunctions.

By comparison, the interface between two regions of dissimilar semiconductor materials is referred to as a "heterojunction." Homojunctions dominate current processes because they are easier to fabricate, but the interface of a heterojunction has naturally occurring electric fields that you can use to accelerate electrons. Therefore, transistors created using heterojunctions can switch much faster than their homojunction counterparts of the same size.

One form of heterojunction that is attracting a lot of interest occurs at the interface between silicon and germanium. Silicon and germanium are in the same family of elements and have similar crystalline structures, which, in theory, should make it easy to combine them but, in practice, is a little more difficult. A process currently being evaluated is to create a standard silicon wafer with doped regions and to then grow extremely thin layers of a silicon-germanium alloy where you need them.

One of the most popular methods of depositing these layers is chemical-vapor deposition, in which you convert a gas containing molecules of interest into a plasma by heating it to extremely high temperatures using microwaves (where plasma is a gaseous state in which the atoms or molecules dissociate to form ions). The plasma carries atoms to the surface of the wafer where they are attracted to the crystalline structure of the substrate. This underlying structure acts as a template, and the new atoms continue to develop the structure to build up a layer on the substrate's surface.

Ideally, such a heterojunction would form between a pure silicon substrate and a pure layer of germanium. Unfortunately, germanium atoms are approximately 4% larger than silicon atoms, so the resulting crystal lattice cannot tolerate the strains that develop, and defects in the structure result. In fact, millions of minute inclusions occur in every square millimeter, preventing the chip from working. Hence, the solution is to grow a layer of silicon-germanium alloy, which relieves the stresses in the crystalline structure, thereby preventing the formation of inclusions (Figure 2).

Silicon-germanium heterojunction devices offer the potential to create transistors that switch as fast as, or faster than, those on GaAs, but that use significantly less power and are based on a robust silicon substrate. Additionally, you can produce such transistors on existing fabrication lines, thereby preserving the investment and leveraging current expertise in silicon-based manufacturing processes.

Diamond substrates

With the drive toward smaller transistors comes the drive toward more densely packed transistors switching at higher speeds. Unfortunately, al-though smaller transistors individually use less power that their larger cousins, modern devices can contain millions of the little rascals, which use a significant amount of power and generate a substantial amount of heat. Thus, although you can see your way to building devices containing more than 100 million transistors by the year 2000, there's a strong chance that such devices would melt into a pool of incandescent slag if you were to use them at their full potential.

And so you come to diamond, which derives its name from the Greek adamas, meaning "invincible." Diamond is famous as the hardest known substance, but it also has a number of other interesting characteristics: It is a better conductor of heat at room temperature than any other material. (It can conduct five times as much heat as copper, which is the second most thermally conductive material known.) It is a good electrical insulator in its pure form, it is one of most transparent materials available, and it is extremely strong and noncorrosive. For all these reasons, diamond would form an ideal substrate material for multichip modules.

A number of methods exist for depositing or growing diamond crystals, one of the most successful being chemical-vapor deposition (the same process you use to create heterojunction transistors). In diamond chemical-vapor deposition, microwaves heat mixtures of hydrogen and hydrocarbons into a plasma, out of which diamond films nucleate and form on suitable substrates. Although technologists do not completely understand the plasma chemistry underlying this phenomenon, polycrystalline diamond films can nucleate on a variety of materials, including metals such as titanium, molybdenum, and tungsten; ceramics; and other hard materials such as quartz, silicon, and sapphire.

Chemical-vapor deposition works by directly growing layers of diamond onto a substrate. A similar, more recent technique, chemical-vapor infiltration, begins with placing diamond powder in a mold. Additionally, you can preform thin posts or columns in the mold and deposit the diamond powder around them. When exposed to the same plasma that you use in chemical-vapor deposition, the diamond powder coalesces into a polycrystalline mass. After you perform chemical-vapor infiltration, you can dissolve the posts, leaving holes through the diamond for use in creating vias. Chemical-vapor infiltration can produce diamond layers twice the thickness of those obtained using chemical-vapor deposition at a fraction of the cost.

An alternative, relatively new technique for creating diamond films involves heating carbon with laser beams in a vacuum. Focusing the lasers on a small area generates extremely high temperatures, which rip atoms away from the carbon and strip away some of the atoms' electrons. The resulting ions fly off and stick to a substrate placed in proximity. Because the lasers are tightly focused, the high temperatures they generate are localized on the carbon, permitting the substrate to remain close to room temperature. Thus, you can use this process to create diamond films on almost any substrate, including semiconductors, metals, and plastics.

Last but not least, in the late 1980s, a maverick inventor, Ernest Nagy de Nagybaczon, invented a simple, cheap, and elegant technique for creating thin diamond films. Nagy's process involves treating a soft pad with diamond powder, spinning the pad at approximately 30,000 rpm and maintaining the pad in close contact with a substrate. Al-though no one fully understands the physics underlying the process, diamond transfers from the pad to form a smooth and continuous film on the substrate. Interestingly enough, Nagy's technique appears to work with almost any material on almost any substrate!

In addition to multichip modules, diamond has potential for a variety of other electronics applications. Because diamond is in the same family of elements as silicon and germanium, it can function as a semiconductor and could be used as a substrate for ICs. In fact, in many ways, diamond would be superior to silicon: It is stronger, capable of withstanding high temperatures, and relatively immune to the effects of radiation (the bane of components intended for nuclear and space applications). Additionally, because of diamond's high thermal conductivity, each die would act as its own heat sink and would rapidly conduct heat away. Some believe that diamond-based devices could switch as much as 50 times faster than silicon and operate at temperatures greater than 500°C.

All of these techniques for forming artificial diamond result in films that are respectfully close, if not equal, to the properties of natural diamond in such areas as heat conduction. Unfortunately, these techniques all result in nanophase structures. (Nanophase materials are a form of matter that scientists discovered only recently, in which small clusters of atoms form the building blocks of a larger structure.) These structures differ from those of naturally occurring crystals, in which individual atoms arrange themselves into a lattice. In fact, many believe that it may be possible to create more than 30 previously unknown forms of diamond using these techniques.

Substrates for ICs require the single, large crystalline structures you find only in natural diamond, but natural gems are relatively small, and today's semiconductor processes are geared to work with wafers 200 to 300 mm in diameter. Unfortunately, there are currently no known materials onto which a single-crystal-diamond layer will grow, except for single-crystal diamond itself (which sort of defeats the purpose of growing the layer in the first place). The only option appears to be to modify the surface of the substrate onto which the diamond layer grows, and many observers believe that this technology may develop in the near future.

Chip on chip

The size of the die increases as a function of the number of transistors it contains, but larger dice result in lower yields. Multichip modules answer this problem by containing a number of smaller, unpackaged dice on the same substrates. However, the intrachip connections linking bare dice on a multichip module are a source of relatively significant delays. One obvious solution is to mount the dice (unpackaged chips) as close to each other as possible, thereby reducing the lengths of the tracks and the associated delays. However, each die can have only a limited number of other dice mounted in proximity on a 2-D substrate. So, you should proceed into three dimensions. Each die is very thin, and, if you mount them on top of each other, you can have more than 100 dice forming a 1-cm cube.

One problem with this chip-on-chip technique is the amount of heat it generates, which drastically affects the inner layers that form the cube. However, you could solve this problem by constructing the dice out of diamond: first, because diamond devices have the potential to operate at temperatures as high as 500°C and, second, because diamond is such a good conductor of heat. Furthermore, the fact that diamond is one of most transparent materials available would facilitate intrachip communication throughout the 3-D cube using surface-emitting laser diodes and phototransistors constructed alongside the standard transistors on each die. Thus, the ability to create consistent, wafer-sized, single-crystal-diamond films would revolutionize electronics as we know it today. If it does prove possible to create such films, diamonds would be not only a girl's, but also an electronics engineer's best friend.

| EDN Access | Feedback | Table of Contents |