|
The Development of Diffusion For Semiconductor Device Fabrication
By John Fairfield (c) SMECC |
The diffusion process was one of the most significant early developments in the manufacture and commercial use of semiconductor devices, such as transistors and diodes. As discussed immediately below, it overcame many of the difficulties of the earlier, laboratory techniques and gave manufacturing a highly reproducible process at reasonable cost. Of all the early developments in the process, diffusion was probably the single breakthrough that most enabled solid state devices to become commercially viable, and has remained an almost universal procedure. Just recently ion implantation, with its even greater control of dopant depths, has replaced diffusion in the fabrication of a most of the newer devices.
As is common knowledge, semiconductor devices are made by introducing small amounts of specific dopants at explicit locations and in carefully controlled concentrations. Such dopants give the semi-conductor the electrical properties required. For example, small amounts of arsenic results in a semiconductor where the electrical current is carried by negative charges, electrons, and is called "n-type" material. Boron as a dopant results in the electrical current carried by positive charges, "holes", called "p-type" material. The location of certain junctions between n and p-type material, for example, those of the emitter and collector junctions, had to be accurately controlled and the distance between them minimized, in order to realize a satisfactory amplification factor (that is, "beta"). Also, the actual distribution of the dopant concentration at the junctions influenced the electrical characteristics of that junction and, therefore, had to be carefully controlled. For example, the sharp gradient of dopant atoms at the junction would tend to increase the diode capacitance, which compromised their value for high frequency applications. Finally, once the dopants were implanted, the junctions had to be located; in order to make contacts. Finally, the fabrication process had to avoid excessive damage to the semiconductor crystalline structure and, more important, to the surfaces.
Historically, the control and reproducibility of the fabrication process was a real problem. The point contact technique worked, to some extent, for simple diodes, but it gave real problems when the relative location of two junctions had to be controlled. Also, there was little control of the structure, dopant concentrations or gradients at the junctions. The reproducibility required for commercial application was impossible. The grown junction, whereby dopants were introduced into the "melt" as the semi-conductor crystal was "pulled", was time consuming and expensive, and it was very difficult to locate the junction with sufficient accuracy to minimize parasitic resistance, capacitance, and other bad effects. The post alloy process mitigated some of the above problems, but it was still hard to control and it introduced defects and precipitates into the crystal structure at the junction, which led to leakage and other problems. Finally, with the demand for smaller devices and very shallow junctions, particularly for fast switching, digital devices, better control was required.
Diffusion afforded a relatively well controlled method of introducing dopants precisely and at shallow depths. Junction depths on the order of one micron were soon accomplished and, equally important, the concentration gradient at the junction could be made steep and controlled. Also, although there are still some problems as discussed at the end of this article, the crystalline structure of the host semi-conductor could be maintained relatively free of additional defects. Diffusion could be combined with certain masking procedures, using photolithography, to minimize the size of the devices, ultimately leading to the complex integrated circuits of today. Electrical contacts could be aligned with the diffusion masks solving the location problem. Also, since the diffusion mask material was an inert, protective substance, usually silicon dioxide, the slight diffusion under the mask minimized, to at least some extent, the formation of troublesome surface states.
Diffusion theory is relatively old and well understood. For the semiconductor industry, diffusion consists of the dopant atoms migrating into the host semiconductor crystal lattice from a source, usually at the surface, by "stepping" through vacant lattice sites, that is, by a substitutional mode. (There are other diffusion modes, e.g. interstitial, but these do not apply to the impurities used for this purpose.) In the diffusion process, the current, or flow, of dopant atoms past a given plane is proportional to the concentration gradient of such atoms; thus, to a first approximation:
current*= D dC
dx
C= dopant concentration
*(in the X direction)
In the above equation, D is assumed constant with respect to the dopant concentration and gradient. This is not quite correct, but the assumption will do for illustration. It is called the diffusion coefficient and is a measure of the ease of diffusion, or ease of movement through the lattice. Qualitatively, we can analyze the physical factors effecting the diffusion coefficient by picturing the process. The dopant atom in a substitutional site vibrates with thermal energy. If, at the extreme of its vibrational displacement, it has sufficient energy to surmount the energy barrier between the lattice sites of the host, and if there is a vacancy at the adjacent site, the dopant atom can step to the next site, thereby migrating in the lattice. Thus, it is seen that the temperature strongly influences the magnitude of D through its influence on the thermal energy of the vibrating atoms and, incidentally, on the vacancy concentration of the host lattice. The magnitude of the energy barrier and the vibrational frequency depends on and is unique to the physical properties of the host lattice and the dopant.
Now the rate of increase (decrease) of the dopant atom concentration in a given, small volume of the semiconductor is proportional to the gradient of the current. In other words, the increase of the dopant concentration is the flow of dopant atoms into the volume minus the flow out of the volume.
Thus:
dC = D d ( dC )
dt dx ( dx )
and we have a second order differential equation. Of course, this equation ignores some second order effects, but will serve as an example. The solution to this equation, of course, depends upon the boundary conditions, can be complicated and is beyond the scope of this article. However, an example of a solution from an initial, given surface concentration would be a negative exponential function with respect to both the direction x and the square root of time. By using the exponential relationship or similar solutions for other boundary conditions, an accurate prediction of the dopant concentrations can be made and a reproducible process can be designed.
Some of the early techniques of diffusion involved depositing a source of dopant on the surfaces of semiconductor wafers by evaporation, painting, spinning, or other technique, and then maintaining these wafers at an elevated temperature for several hours in an inert gas. Usually this was a dynamic process with the gas flowing at some controlled rate. As time went on, the dopant was often added to the ambient gas by bubbling the gas through a liquid compound of the dopant, by heating the dopant or some compound thereof in the gas, or by introducing a small amount of a gaseous compound into the ambient. This approach was more reproducible and cleaner that the "paint on" method described above, and it became a standard process for many years. However it introduced the basic processing problem of impurities in the ambient, which caused surface problems. Moisture was always a troublesome contaminant, and some dopants, especially phosphorus (for n-type) and gallium (for p-type) would combine with the OH ion and attack the exposed surfaces of the semiconductor. One early idea was to add about 10% hydrogen to the ambient, hopefully to "getter" the oxygen. This was never satisfactory. This problem was solved, by the development of much better techniques to purify gases; and, more important, to measure very small amounts of impurities. It was also solved by better hardware. Initially gas systems included high quality vacuum tubing of some rubber or plastic component. There was always trouble with these materials, which, in retrospect, is not very surprising; and, gradually, all the systems went to high purity quartz and stainless steel only, even for the diaphragms of pressure regulators. No rubber or plastic. The joints between the stainless and quartz were a problem, but a ball joint approach was developed with no neoprene O-ring.
Another technique was to seal the wafers in a quartz capsule with a dopant and heat the entire capsule for several (up to twenty-four) hours. This process was more costly, but it greatly alleviated the impurity problem, and afforded better control of the surface concentration of the diffusant. Subject to the purity of the quartz, which can be made very pure compared to Pyrex and other glasses, the system could be kept pure. Usually the diffusant source was a powdered form of an alloy of the diffusant and the semi-conductor, which resulted in a very well controlled vapor pressure of the diffusant for the longer diffusion times typical of this technique. Finally, the powdered semi-conductor resulted in a vapor pressure of the semi-conductor itself, which tended to protect exposed surfaces. In other words, this was an equilibrium process. All of these techniques have survived for many years.
Although the diffusion process was a significant development, there are certain limitations and problems that must be considered. First, crystal defects, such as dislocations, twin planes, etc., can and has been shown to alter the diffusion coefficient through the abnormally high vacancy concentrations that are normally around crystalline defects. This effect is difficult to control, and can lead to shorted devices because of a very fast but highly localized diffusion. This caused the famous "pipe" problem where the emitter and collector became a dead short. Also, the diffusion of some dopants was abnormally high along twin planes or surfaces between the semiconductor and the masking material. Diffusion under the mask became a particular problem for MOS devices where a precise lateral dimension was required to control the gate width. This problem was significantly alleviated only recently through the development of ion implantation.
The process needed to be carefully controlled to minimize damage to the crystals. Since the diffusion temperatures approached the melting point of the semiconductor, the wafers had to be supported so as to minimize the stress. Otherwise, excessive crystal defects would form. To minimize unwanted impurities, usually high purity quartz was used to enclose the diffusion systems, as discussed above. However, this was always a problem, especially for cases where a fast diffusing atom existed that would compromise the electrical performance of the devices. Sodium gave big problems and, at times, came from the fingers of the technicians performing the process.
The dopants themselves could strain the host lattice and cause problems associated with defects. For example, phosphorus was a much used n-type diffusant in silicon, but phosphorus was a poor fit to the silicon lattice, and high concentrations would result in dislocations, which effected both the electrical performance and the diffusion coefficient. Conversely, arsenic afforded a better fit and a more classical diffusion profile, but its rate of diffusion was slower and not always compatible with other diffusants. Nothing much could be done about this problem, and it was minimized by clever design of the processor, and carefully choosing the diffusants.
Another problem, though less serious, resulted from the fact that the rate of diffusion is influenced by high concentrations of the dopant and by high gradients. This is caused by an effective electric field, that occurs in semiconductors around sharp gradients of junctions, which electric field influences the migration of the dopant atoms since they are usually in the ionic state. Finally, diffusion is influenced by impurities other than those of primary concern, which results in "cooperative diffusion". Thus the diffused impurities for one junction, e.g. those for the emitter, can influence the diffusion of those atoms for a different junction, e.g. the collector. As stated above these effects have caused less problems since they are reasonably predictable and can be allowed for in designing the process.
In summary, diffusion is a process that became important in the early days of the semi-conductor industry. It has remained a very significant processing technique. -JF
|
About John Fairfield
John Fairfield received a Bachelor’s Degree in Physics from Dartmouth College in 1954. After three years of service in the United States Navy, he attended and earned a Master’s Degree from the University of Chicago in 1958. Mr. Fairfield then worked for IBM and was affiliated with the Product Development Laboratory at Poughkeepsie and East Fishkill, New York. He left IBM in 1969 to help found Semiconductor Electronic Memories, Inc., a small integrated circuit memory manufacturer in Phoenix, Arizona. He has been a member of the American Electro-Chemical Society and the American Physical Society.
Mr. Fairfield is now managing commercial property including industrial facilities, as co-founder and general partner of the Fairfield Company, Phoenix, Arizona.
|
| Add Picture of John and FIX THE FORMULAS WITH SCANNED
IMAGES! |
|
