Fullscreen HTML5Med Res (??MB)HTML4

The History of Transistors Last month, I described the invention of transistor technology and some of the early techniques used which were not well suited to mass production or high performance. This article continues where that one left off, covering the rapid progress in the 20 or so years between the first commercial transistor production and the development of manufacturing techniques that are still in use today. Part 2: by Ian Batty T he first article in this series described the ‘hand-made’ phase of transistor construction. Although some processes were automated, they were very much made one at a time. That’s still true of the first few techniques we’re about to look at. Alloying, for example, required each transistor’s indium dots and base slice to be individually assembled for loading into the alloying furnace. The breakthrough came with the application of photolithography. Combined with gaseous diffusion, 38 Silicon Chip this provided all stages of fabrication apart from terminal and lead attachment. This meant that manufacturers could automate the manufacturing process and apply batch processing to yield many devices from one feedstock wafer/slice, as we shall soon investigate. Alloyed-junction transistors Grown junction technology was demonstrably superior to point-contact but could not yield a base of sufficient thinness for operation much above Australia's electronics magazine 1MHz. The Regency TR-1’s designers were forced to use an intermediate frequency (IF) of only 262kHz to get reasonable gain. The alloyed-junction transistor was invented by John Saby at General Electric, with similar developments undertaken by Jacques Pankove at RCA. Inventorship had to be established in the US courts, as RCA had filed on Pankove’s work one day ahead of General Electric in June 1952. They were initially PNP types and commenced with a wafer of N-type siliconchip.com.au Fig.22: alloyed-junction transistors were made by adding indium pellets or ‘dots’ on the surface of the N-type (doped) silicon base, then heating the assembly in an oven until the P-type indium formed an alloy with sections of the base. germanium, typically doped with antimony (becoming the base). Some details of production are set out in Pankove’s patent at https://patents.google. com/patent/US3005132 The junction transistor was created by alloying emitter and collector dots onto the base slice at high temperatures. This design was reliable and economical to produce. The famous OC70/71 and OC44/45 series used in the late 1950s and early 1960s were all alloyed-junction types. Alloyed-junction transistors worked at moderately high frequencies – up to about 15MHz for the OC44. Point-­ contact transistors were still in limited use, as their highest operating frequencies extended to around 300MHz. Current flow in a PNP transistor originates at the emitter, crosses the emitter-base junction, diffuses through the base, then crosses the base-­collector junction. The slowest movement is within the base, so the first area for improvement was to make the base as thin as possible. The principal problem with this was in the alloying process. A typical transistor began with the N-type base slice having P-type indium ‘dots’ for the emitter and collector placed on it. The assembly was then heated to the melting point of indium, below germanium’s, forming a eutectic alloy that combined the indium into the germanium – see Fig.22. A practical transistor has a base thickness measured in micrometres; the base thickness is exaggerated in Fig.22 for clarity. You can see photos siliconchip.com.au of ‘delidded’ germanium alloyed-­ junction transistors in Figs.23 & 24, in which the alloyed indium dots are clearly visible. The molten indium penetrated the germanium base area from either side. The aim was to alloy the emitter and collector as closely together as possible without ‘shorting out’ the base region. An article extracted from the January 1961 edition of Mullard Outlook (siliconchip.com.au/link/abbi) describes just how laborious and handmade the OC71 and its fellows were. The Outlook describes how the collector sites were alloyed first, then the base slice removed from the furnace, turned over, the emitter sites placed and the entire assembly re-alloyed to complete the transistors. Raytheon solved the multi-pass problem by inserting the emitter dot into a recess in a small graphite ‘boat’, then placing the base slice, then the collector dot on top of the base to complete the ‘sandwich’. The entire assembly then went through the alloying furnace, creating the transistor in a single pass. In practice, base thicknesses of Fig.23: a delidded transistor from an IBM 1401 computer. Early versions of that computer used standard alloyedjunction germanium transistors, while later versions used faster, diffused ‘drift field’ transistors. Source: Marcin Wichary, USA (CC BY 2.0) much less than about 0.5 thou (0.0005in or about 0.013mm/13µm) proved difficult to produce reliably. Philips’ commonly-used OC44, with its cut-off frequency of around 15MHz, was bettered only by RCA’s 2N1308 at 30MHz, representing the state-of-theart for alloyed junction transistors. Many OC45s were simply OC44s that had not met the OC44 specifications and were marketed as perfectly good devices with a high-frequency cut-off of only 6MHz. You’ll see radios of the time with an OC44 converter and OC45s relegated to the IF. Once production was running smoothly, manufacturers concentrated on improving important parameters such as the power rating, maximum voltage, high-frequency response, temperature stability and noise. Many valves did not have frequency ratings, and some turned out to be suitable for use at much higher frequencies than intended, such as the 6BE6 pentagrid operating up to 108MHz in FM tuners. On the other hand, junction transistors universally have maximum frequency specifications. This is useful Fig.24: a small-signal alloyed-junction germanium transistor suitable for audio use. Note how the base and collector leads are insulated from the metal can while the emitter, hidden under the wafer slice, connects directly to the can for shielding. Image copyright: Jack Orman, www. muzique.com Australia's electronics magazine April 2022  39 for designers but also points to the intensive design and development efforts that now allow transistor operation up to and beyond 500GHz. Raytheon’s landmark 8TP portable radio was the second all-transistor radio behind the Regency TR-1. But unlike the TR-1, judged by Consumer Reports in April 1955 as a “toy that didn’t come at a toy-like price”, its performance was quite credible. The 8TP, like most Australian radios built in the 1950s, uses PNP types in the RF/IF section. This is a good indication of alloyed-junction construction as alloyed-junction types are most easily made using indium-alloying. It’s possible, but difficult, to create NPNs using alloying. Diffused construction Alloyed-junction transistors were incapable of working much above 30MHz, a limit easily surpassed by the less-reliable point-contact technology. With the physical base thickness restricted to a minimum of about 10μm using alloying techniques, designers turned to the question of current-­ carrier speed across the base. Four solutions were found. The first was to produce a graded doping concentration from emitter to collector, the ‘diffused junction’. The second was to alloy using the collector as the substrate; the base was diffused into the collector slice, and the emitter alloyed into the base ‘surface’. The third was to chemically etch a very thin base area for emitter and collector deposition. The fourth was to use diffusion to fabricate the base and emitter over the collector substrate. 1) Graded doping Charge carriers must diffuse across the base-collector junction, and a uniform doping concentration does not give the fastest transit time. If the doping concentration is modified across the thickness of the base, charge carriers experience less recombination and get a comparative ‘boost’ in their slow diffusion towards the collector. Light doping near the base-­collector junction also reduces the effective capacitance in that area, thus reducing collector-base feedback capacitance. RCA’s drift field process, proposed by Herbert Kroemer in 1953, was put into production by 1956. 40 Silicon Chip Fig.25: graded doping allowed for lighter doping in the collector area of the substrate, forming a ‘drift field’ transistor that accelerated electrons/holes more effectively, therefore improving high-frequency operation. Rather than doping the entire base slice at manufacture, just one side of the base was exposed to a doping gas in a furnace. This caused a high doping concentration on the exposed side, and progressively weaker doping as the doping gas diffused through the germanium base slice. The resulting doping gradient, shown in Fig.25, allowed higher-­ frequency operation. This method still relied on physically thin and fragile base slices, and offered no means of reducing the active base thickness. Drift transistors such as RCA’s 2N247 offered cut-off frequencies up to 60MHz. Even with strict control of the alloying process, 60MHz was probably the practical limit for such construction. While this technique went no further, it demonstrated that uniform doping of the base could be dispensed with, and hinted that the transistor might be fabricated on one side of the substrate. 2) Collector substrate Philips, meanwhile, had extended the concept of diffusion into the germanium slice. Rather than a two-sided approach fabricated on a base substrate, they began with a relatively thick and mechanically robust collector slice. The first approach was to create a thin doped layer to diffuse the base into the collector substrate using diffusion. An emitter ‘dot’ was placed onto the base surface and alloyed into the base layer, as Fig.26 shows. The initial design used a contacting ring to connect to the base diffusion, superseded by an alloyed base connection. The base dot was of the same Fig.26: gas diffusion allowed a very thin surface layer on the wafer to be doped to N-type while the bulk of the silicon remained P-type. The emitter was then alloyed on top. This thin base layer provided even higher frequency operation. Fig.27: a refinement of the scheme shown in Fig.26; here, a base dot is alloyed along with the emitter. The base dot is N-type material on top of the N-type diffused layer, so it doesn’t form a semiconductor junction; just a convenient electrical connection to the base layer. Australia's electronics magazine siliconchip.com.au Fig.28: another germanium audio transistor, apparently a diffusion type as no alloyed dots are visible, and the wafer slice is so small that it’s mounted on a stamped steel base to keep the bond wires short. Image copyright: Jack Orman, www.muzique.com polarity as the existing base (N-type in the OC169~171), the emitter dot of P-type. Alloying the base dot simply made electrical contact with the base layer, but the emitter dot would alloy into the base, forming the emitter-base junction, as shown in Fig.27. This principle is also known as the Post-Alloy Diffused Transistor (PADT), as the alloying follows base diffusion. In 1957, J. R. A. Beale reported experimental production with operating frequencies up to 200MHz. In full-scale production, devices such as the OC169~171 could operate at 100MHz. The AF118 RF/Video amplifier boasted a cut-off frequency of 175MHz. Production spreads still existed: Fig.29: by acid etching the base layer to make it as thin as possible before adding the emitter and collector, the effective base could be made thinner, thus speeding up current flow across it. This resulted in more fragile transistors. transistors were graded for performance at 100MHz, with the best-­ performing OC171s intended as RF amplifiers in the 88-108 MHz FM band. The OC170 and OC169 were recommended for converter and IF amplifier service, respectively, and came from the same production lines. Further development yielded some impressive results, with the AF186 posting a cut-off frequency of 820MHz. A UHF tuner design from 1967 gave a gain of 22dB with a noise figure of 10.5dB at 860MHz. This was already superior to competing valve designs, which principally used a valve local oscillator and solid-state diode mixer, but no RF amplifier. Microwave valves such as the discseal 6BA4 or the ceramic 7077 gave good RF amplifier performance, but were not considered practical in mass-produced consumer electronics. Even with this level of performance, the days of alloying were numbered. The AF186 came in two varieties, pre-amplifier and mixer-oscillator, implying that significant manufacturing variabilities still existed. Also, the entire alloying process was illsuited to high-yielding mass production demands. Some online sources describe the previous alloyed-junction construction as “diffused”. Early confusion over whether the indium-germanium consolidation was a diffusion or an alloying process was finally resolved by John Saby in 1953 (see siliconchip. com.au/link/abbj). A true diffused construction relies on diffusing a doping gas’ doping concentration into the depth of the base (or collector) at high temperatures when the base is manufactured. Against this, alloying relies on the eutectic process, where the alloy’s melting point is lower than that of either constituent, just as tin-lead solder melts before either tin or lead. But you can argue that an alloy also sees mutual diffusion of one part into the other. Saby recognised this, accepting ‘alloying’ over ‘alloy-diffusion’ for the alloyed-junction process to distinguish between the older dot on a slice process and the newer gaseous atmosphere processes being developed (see siliconchip.com.au/link/abbk). 3) Base-substrate etching Alloyed designs already used the thinnest practical base of uniform thickness, but only the section directly between emitter and collector needs to be as thin as possible. Why not use a suitably thick base substrate for mechanical strength, then thin out the area where the emitter and collector will be formed? Philco (https://en.wikipedia.org/ wiki/Philco) invented the surface barrier transistor (SBT) and used this technology to build the world’s first solid-­ state computer in 1957, the S-2000 Transac. Arthur Varela used precision etching to chemically erode the base slice, forming a ‘well’ on either side. Fig.29 shows how the etching created the thin base region, then transformed into a plating process, with the emitter and collector regions being plated onto the base surface. Devices such as the 2N240 could operate up to 30MHz. See also Fig.30. Fig.30: a page from US patent 2,885,571 shows how acid etching makes the base extremely thin, speeding up the transistor. siliconchip.com.au Australia's electronics magazine April 2022  41 Fig.31: the diffusion process for making diodes. Variations on this process could be used to add a third layer for making transistors. Impressive as the SBT was, it still relied on highly precise manufacturing that was not easily adapted to automated, high-volume mass production. Additionally, the very thin base was still mechanically weak, so the device was prone to damage from vibration or shock. The similar micro-alloy transistor (MAT) enjoyed a brief appearance, especially in England, where Technical Suppliers Limited and Clive Sinclair’s Sinclair Radionics offered these devices. Notably, the TSL catalog shows MATs with cut-off frequencies of 75MHz, but a competing alloy-­ diffused device with a cut-off of 400MHz. 4) Micro-alloy diffusion If diffused-base technologies such as the drift-field allowed operation up to 60MHz, why not apply diffusion to a very thin base? Would this improve the performance of the surface-barrier design? Philco engineers addressed this problem, applying etching techniques to a diffused base: the micro-­ alloy diffused transistor (MADT). As noted above, the entire base does not need to be extremely thin, only the section between emitter and collector. The 2N502A MADT could oscillate up to 500MHz. Micro-alloy diffusion works by using diffusion techniques to create a doping gradient through the base that promotes rapid charge motion from emitter to collector. ‘Wells’ are then etched into the base slice to form the thinnest possible base section between emitter and collector. Finally, the emitter and collector surfaces are ‘plated’ into the base wells, ready for lead attachment. All-diffusion techniques The previous fabrication methods (especially those using etching) could 42 Silicon Chip not make good use of high-volume, automated manufacturing techniques. Two solutions were found: mesa and planar processes, each allowing hundreds of transistors to be made on a single semiconductor wafer/slice in one go. That slice was then cut apart to yield individual transistor ‘chips’ ready for testing and packaging. Also, if devices other than transistors could be fabricated on a wafer and interconnected, it would be possible to create many individual, functional circuits on a wafer. Finished circuits could be tested in place, the good ones cut and packaged, and the rejects discarded. But that would come a bit later. The key to this revolution was photolithography (https://patents.google. com/patent/US2890395A), a refinement of the photographic techniques used to make printed circuit boards (PCBs). Paul Eisler’s 1943 patent application (GB639178) for the PCB became the basis for proximity fuse design in anti-aircraft shells. Post-war declassification was followed by Moe Abrahamson and Stanislaus Danko’s patent, granted in 1956 (https://patents.google.com/patent/ US2756485A) – see the last page. Mesa fabrication Working at Texas Instruments, Jack Kilby developed the monolithic (‘single stone’) mesa process. Mesa construction (named for flat-topped tablelands of south-western USA) began with a substrate slice of doped silicon, let’s say N-type. The substrate (commonly known as a wafer) was placed in a furnace and exposed to a P-type doping gas. This created a P-type layer over the entire surface of the substrate, forming a single diode junction (see Fig.31). At this point, the slice could be cut up into chips, each being one P-N diode. Photolithography The photolithographic process allows the creation of individual Fig.32: this is an example of how a circular slice of silicon (or germanium) crystal could be made into many separate diode dice using photolithography and acid etching. Australia's electronics magazine siliconchip.com.au Fig.33: the Mesa process was an early photolithographic transistor production method with important advantages. doped ‘islands’ on the slice rather than a continuous doped surface. Fig.32 shows an example mask applied to a germanium/silicon wafer. The key to the process is a photosensitive resist. This chemical responds to ultra-violet light by hardening and adhering to the surface it is applied to. Exposed resist will remain in place during processing, while unexposed resist (covered by opaque parts of the mask during UV exposure) is easily washed off to permit the doping atmosphere to diffuse into exposed areas of the wafer. Beginning with a wafer that has been processed to create a single large P-N junction, the slice is resist-coated, masked and exposed to ultraviolet light. The light not obscured by the mask passes through and hardens the resist layer. The unexposed resist is then washed off, leaving a protective pattern over the slice. This process is shown in Fig.33. The slice is then exposed to an etching acid so that the unprotected areas of the slice are dissolved, removing the P-type layer in the exposed areas. Precision control results in the desired P-type ‘islands’ over the surface of the N-type substrate. The etching creates side trenches, separating each ‘island’ from its neighbour, and giving the distinctive ‘Mesa’ profile. Finally, the resist layer is removed from the entire slice, and it is cut up to yield Mesa diodes. This process can be automated, with the slice never leaving the production line’s controlled atmosphere. This eliminates the possibility of surface contamination, giving much higher consistency and reliability. The example mask in Fig.32 would produce 160 diodes in one production run. In reality, 1969s standard siliconchip.com.au two-inch wafer/slice produced hundreds of diodes. Making mesa transistors The process for making transistors is similar, but naturally, it is a little more complicated. First, a fresh substrate is placed into a furnace and the entire surface is exposed to a doping gas, making a single P-N junction, as before. The next stage is to take the entire slice and coat it with resist, just as for the diodes. But this mask contains smaller holes – each one overlaying a part of the previously-doped P-type base material where the exposed area is to become the emitter of a transistor. UV exposure hardens the resist layer, and the unexposed portions are washed off. The slice is exposed to an N-type doping gas in the furnace, changing the exposed P-type silicon base areas to N-type doping, creating the small emitters. We now have an N-P-N structure. Finally, the edges of the useful area are etched to isolate each transistor from its neighbour (similar to Fig.33) and ensure that the N-type substrate is isolated from the collector around the edges – just like the diodes. The transistor has been made with two diffusion processes – first the base, then the emitter into the formed base area. Thus, this type of transistor is known as double-diffused. Mesa’s double-diffusion can be conducted in a single pass through the furnace. Beginning with an N-type substrate, aluminium vapour (a light, rapidly-diffusing acceptor impurity) can be made to diffuse and create the P-type base, a layer only 0.0001 inches or about 2.5µm thick. That’s a bit less than four wavelengths of visible red light. Simultaneously, and in the same atmosphere, a more slowly-diffusing antimony donor impurity penetrates less deeply, following the aluminium diffusion, overcoming and reversing the aluminium’s acceptor doping. This creates a shallower N-type emitter layer extending from the surface into the base region. Consider the astounding precision of control needed for this process and the fact that the substrate stock’s purity is measured in parts per billion. That’s why the Mesa process (and its successor, planar, described below) took so many thousands of laboratory hours, so many millions of dollars and so many years to reach perfection. Fig.34 shows a simplified example of a NPN transistor of Mesa construction. Notice the etched side ‘trenches’ that isolate the collector-base junction. Fig.34: this is how a transistor made using the Mesa process looked when finished. The name comes from its distinctive shape, like a desert mesa. Australia's electronics magazine April 2022  43 Input 1 Output 1 B+ Earth Output 2 Input 2 Mesa/double-diffused construction gives great improvements in yield: many individual devices can be fabricated on a single germanium or silicon slice, and the high-precision nature of photolithography allows for the creation of much smaller individual devices. Although the illustration does not show it, Mesa devices can also use epitaxial construction, described in the next section. A parallel development allowed robotically-controlled, microscopic probes to examine and test each finished device on the slice. Faulty or below-standard devices would be Fig.35: a photo of the first commercial integrated circuit, Fairchild’s µL902 flip-flop from 1961. It was made using a similar process to the epitaxial planar technique, with more complex lithographic masks to create the four transistors and their interconnections. Source: Fairchild recorded in computer memory and rejected once the slice was scribed and broken up to produce individual devices. Mesa technology drove costs down and yielded devices with greater reliability and performance figures. Texas Instruments advertised their germanium 2N623, with a maximum oscillating frequency of 200MHz, in July 1958. By March 1959, TI’s 2N1141 could operate to 750MHz. While this performance is about equal to the best alloy-diffused transistor, the process delivers higher yields and is therefore more economical. Although surpassed by its planar successor for high-frequency use, Mesa technology is still widely used for high-power transistors. At about this time, Jack Kilby pioneered integrated circuits by fabricating several devices onto one germanium “chip”, forming a simple digital circuit. Those devices still relied on fine interconnecting wires between the devices. He was awarded the Nobel Prize for Physics in 2000. Despite Kilby’s invention being regarded as ‘the first’ integrated circuit, Kilby does not have absolute priority (https://patents.google.com/patent/ US3138743A). Harwick Johnson filed a patent in 1953 for an analog phase-shift oscillator in a “unitary body” that we would recognise as an integrated circuit. Johnson’s device did not rely (as Kilby’s did) on manually-placed interconnecting wires to complete the device (https://patents.google.com/patent/ US2816228A). Six months later, Robert Noyce, one of the “Fairchild Eight”, perfected integrated circuit design by vapour-depositing metallic wiring interconnections over the chip surface, creating a device that could be Fig.36: the planar epitaxial diode manufacturing process, which can be considered the direct predecessor of many Fig.37: epitaxial planar transistor manufacturing starts with the output of the diode process and repeats essentially the 44 Silicon Chip Australia's electronics magazine siliconchip.com.au made entirely automatically. Noyce (in contrast to Kilby) used silicon, starting the IC revolution that has given us everything from supercomputers to smartphones with cameras (see Fig.35). Planar transistors The final phase of development coincided with the implementation of fully automated fabrication. As mentioned in part one, ideally, the base of a transistor should be as thin as possible for the highest frequency of operation. But the base still needs an electrical contact made to it, and such contacts have practical size limits. Mesa technology used edge-etching to define the edges of the junctions, potentially exposing the junctions to contamination. The collector should also ideally have excellent conductivity (for the least possible electrical resistance and best high-frequency performance), but this demands doping too heavy for practical devices, as it gives very high collector-base capacitances. A very thin collector with light doping would give the desired low resistance and low capacitance. But, as this would be too fragile for practical devices, some compromise was always forced on designers. Epitaxial planar Howard Christensen and Gordon Teal’s 1951 patent solved the thickness/resistivity problem by showing how to grow a very thin and lightly doped semiconductor layer over a more heavily-doped thicker substrate (https://patents.google.com/patent/ US2692839A). Jean Hoerni’s patent of March 20th 1962 demonstrated an advance on Mesa technology: epitaxial planar manufacture, using Christensen and Teal’s epitaxial process (https://patents.google. com/patent/US3025589A). The epitaxial (“arranged around”) layer has an identical crystalline structure to the substrate, but can have any degree of doping concentration and even the opposite doping type. This remains essentially the state-of-theart for semiconductor manufacture to the present day. Like the Mesa process, Hoerni’s technique uses double-diffusion: base into collector, emitter into base. Fig.36 shows the manufacturing of diodes with this technique. A lightly-­ doped N-type epitaxial layer is grown over the N-type substrate using gaseous diffusion – basically, a form of controlled condensation. The substrate is coated with photo-­ resist, then masked. Ultraviolet light shines through the mask, hardening the exposed resist layer. The unexposed resist is washed off, and the slice is exposed to a P-type doping gas to form a diode. After washing off the exposed resist, the anode and cathode connections are made, and the diode is complete. This gives the desired thin, lightly-­ doped layer needed for when the process continues to manufacture transistors (as shown in Fig.37). It has a low-capacitance junction in contact with a sturdy and highly conductive layer below. Beginning with the diode structure, the slice is resist-coated, masked and UV-exposed to leave part of the existing P-type diffusion unprotected. After washing off the undeveloped resist, the slice is exposed to N-type doping, which diffuses into the base. This gives the N-P-N structure for a transistor. Finally, the entire surface is oxidised to form a silicon dioxide protective surface. This oxidation phase makes epitaxial planar manufacturing modern transistor manufacturing processes. same steps to add the third (emitter) layer. siliconchip.com.au Australia's electronics magazine April 2022  45 Fig.39: a Fairchild ► epitaxial transistor die. The star shape conferred some performance advantages over a circle. Source: Fairchild Fig.38: a finished epitaxial planar transistor. The silicon dioxide (SiO2) layer on top insulates the transistor and provides a barrier against moisture and contaminants. This allows the transistor to be housed in a low-cost plastic package. unsuited to germanium devices: germanium oxide is soluble in water and fails to form a protective layer. A final masking-etching step produces small apertures in the SiO2 mask to allow metallisation for emitter and base contacts, as shown in Fig.38. Alternatively, it’s possible to diffuse directly through the SiO2 layer to make contact with the desired areas. The active device now has a thick substrate for strength with low resistance, a collector layer with the desired lighter degree of doping needed for transistor action and low collector capacitance, a diffused base layer and an emitter layer diffused into the base. This leaves a base layer of the desired thinness for the intended maximum operating frequency. The collector contact is made to the collector substrate. Individual transistors are robotically tested, the slice is broken up, ‘good’ chips are selected and encapsulated with connecting leads attached. The SiO2 passivation layer’s robustness makes encapsulation in cheap epoxy resin possible; germanium Mesa devices needed metallic casings to guarantee hermetic sealing. Fairchild released their 2N709 in March 1960, with a maximum operating frequency of 600MHz. The 2N709A pushed this to 900MHz. A final advantage of photolithography is the ability to create devices of any geometry. Simple circular cross-section devices do not give the best RF performance, especially at high power levels. Mesa and planar devices can use complex geometries unobtainable by previous processes. Fig.39 shows a microscope photograph of a star-shaped Motorola epitaxial 2N2222 transistor die. 46 Silicon Chip Note that all illustrations in this (and the previous) article significantly exaggerate the base thickness (and emitter, in some cases). In practice, base thickness is measured in micrometres. Several fine publications have attempted to give some impression of the true scale of fabrication, but the results are difficult to interpret because of their attempted fidelity. Fig.40 is an original alloy-diffused diagram from Mullard’s Reference Manual of Transistor Circuits (at approximate full size here), illustrating the problem of accurate visualisation. Silicon’s advantages over germanium Germanium’s relatively low melting point of 940°C made it the material of preference for point-contact and early junction transistors, but it has fallen into disuse for several reasons. Silicon NPN and PNP annular epitaxial transistors ... Designed for complementary high-speed switching applications and DC to 100 mc amplifier applications. First, the collector-base junction proved to have significant reverse (leakage) currents even with no forward base bias applied, meaning that the transistor could never be truly cut off. These leakage currents worsened at elevated temperatures. While this might be tolerable in diodes, leakage in transistors could lead to rapid increases in collector current. Such increased current causes increased heating, causing increased leakage, causing increased heating... this is thermal runaway. Think of a 1960s car radio in the middle of summer; cabin temperatures could easily exceed 50°C. In a power transistor, the device can rapidly increase its current from its desired bias value (of milliamps) to a current of many amps and be destroyed by overheating. Circuit designs must stabilise the transistor against such current variations. Also, Fig.40: this diagram from Mullard attempts to show the features of a transistor at actual size, making some of the details such as the base layer hard to see, as they are so thin compared to everything else. Source: Mullard Reference Manual of Transistor Circuits, 1960 Australia's electronics magazine siliconchip.com.au germanium junctions can only operate to about 70~90°C while silicon devices are commonly specified up to 200°C (although 175°C is more typical). One early car radio was notorious for its OC72 germanium output transistors overheating and being destroyed. Silicon junctions exhibit much lower leakage currents, giving better performance, especially at high power levels and high temperatures. Thermal runaway must still be considered, but it is far less of a problem with silicon. Also, silicon is much more plentiful. If you’ve ever rinsed sand out of your bathers, you’ve rid yourself of the raw material for thousands of transistors and many microprocessors! Germanium is a rare element. Back in the 1950s, germanium ore was so scarce that one transistor manufacturer was forced into recovering germanium from the flue ash of power stations. It’s still scarce, as expressed by its 2018 price of some $2600/kg, with pure silicon costing as little as $50/kg. Silicon’s advantages are somewhat counterbalanced by its much higher manufacturing temperatures (almost 1400°C) and the difficulties of adapting germanium manufacturing techniques. Impurity doping methods that worked well with germanium had to be modified. For this and other reasons, early silicon transistors performed poorly at high frequencies. Parallel advances in mass production techniques gave them an initial cost advantage, however. Once begun, silicon processing developed rapidly. Silicon did offer one major manufacturing advantage over germanium: the oxide of silicon (basically, glass) is highly insulating and resistant to liquid or gaseous contamination. Silicon devices could be ‘finished’ with a final layer of SiO2, creating localised hermetic sealing, greatly improving reliability and allowing encapsulation in cheap epoxy resins. Germanium dioxide lacks these properties. Early silicon transistors offered little performance improvement over germanium types. Still, manufacturers focused their efforts on improvement, finally offering devices superior in every parameter except for base bias voltage: about 0.6~0.7V for silicon compared to 0.15~0.25V for germanium. This single advantage was insufficient to outweigh germanium’s disadvantages. Gordon Teal somewhat mischiesiliconchip.com.au vously sprang TI’s first silicon transistors on an amazed IRE National Conference in Dayton, Ohio in 1954. TI’s silicon devices rapidly supplanted germanium types. This advance contributed to the total collapse of Philco’s and Raytheon’s transistor divisions, as they could not rapidly shift from germanium feedstock and processes to silicon. Growing from a humble electrical company founded in 1892, Philco became a supplier of batteries for first-generation electric vehicles in 1906 and was the creator of the first all-transistor portable television and the world’s first all-transistor computer. Despite landmark aerospace and computing contracts, Ford bought out Philco in 1961 and ceased to exist as an independent corporation. By the 1960s, silicon had become the dominant material for semiconductor fabrication. A final ‘wrinkle’ in the story is that, for alloyed germanium, the PNP structure is optimal, but for silicon planar, it’s NPN! Now we know why all those germanium Philips transistors in our junk boxes need a negative battery supply, and why their silicon cousins need the opposite. Simply, it’s all about doping. Why not tetrode construction? All modern transistors, aside from dual-gate FETs, are triodes (ie, they have three terminals). Compare this to valves where triodes gave way to tetrode and then to pentodes in amplifying circuits. Junction transistors are built from three distinct layers – emitter, base, collector – and the device current originating from the emitter must pass through the base layer to arrive at the collector. This current path is equivalent to that of a thermionic triode, where the cathode current must pass through the grid to reach the anode. Field-effect transistors have no ‘intermediate’ electrode between the source and drain. Source current passes directly along the channel, but is influenced by the electrical bias field from the gate. Part of the reason behind the popularity of tetrode and pentode valves has to do with two principal limitations of thermionic triode operation. First, the proximity of the anode to the grid created significant capacitance that limited triode performance at high frequencies. This problem was eventually solved by the addition of the screen grid, creating the tetrode. A valve tetrode’s electron stream simply passes through the screen’s thin wire helix. Such screening construction proved impractical with junction transistors, so the problem of collector-base capacitance could not be eliminated. Designers had to work to reduce the existing collector-base capacitance to the lowest possible value instead. Tetrode transistors were made, but they still possessed only two junctions, as shown in Fig.41. The extra connection went to the opposite side of the usual base contact. Applying a repelling bias to the B2 connection forced charges away from that side, narrowing the flow of charge carriers. In effect, the active junction’s area could be controlled electrically. Fig.41: dual-base or ‘tetrode’ transistors were created early on to overcome some limitations of the transistor technology of the time. But since then, other ways have been found to improve the transistor’s performance. So except for a few dual-gate Mosfets or Mosfets with separate substrate connections, pretty much all modern transistors have three terminals. Australia's electronics magazine April 2022  47 This improved high-frequency performance; the amount of internal resistance in the electrically-smaller base slice was reduced, as was the base-­ collector capacitance. Tetrode transistors enjoyed a brief period of implementation but were overtaken by improvements in ‘triode’ designs, and are now obsolete. The second reason for using the screen grid in valve amplifiers was to achieve much higher voltage gain than triodes could provide. That was a discovery made after the primary aim of reduced anode-grid capacitance had been achieved. Thermionic triode voltage gains are limited by the fact that the anode voltage affects anode current; lower anode voltages mean lower anode currents. The ratio of grid control of anode current to anode control of anode current is the valve’s amplification factor, its mu (µ). Valve voltage amplifier triodes have µ values from around 3 to 100, with the ‘negative feedback’ effect of anode voltage forbidding any higher practical gain. Even early tetrodes gave µ values of several hundred or more, and the 6AU6 pentode can give a µ as high as 5000. The screen grids of tetrode and pentode designs eliminate the ‘feedback’ effect of anode voltage; anode current remains virtually unchanged with changing anode voltages. This can be expressed in either of two ways: the valve appears in-circuit as a very high resistance, acting as a constant current device, and the characteristic curves are virtually flat above some 20% of normal operating voltage. Transistors, both junction and field-effect, all exhibit ‘pentode’, constant-current characteristics as amplifiers. Changes in collector (or drain) voltage have little effect on current. Transistor output resistances or impedances range typically from tens to thousands of kilohms. Such characteristics mean that, even if a screen layer could be added, it would deliver little extra in the way of gain. Mullard detail a (triode) OC70 circuit with a voltage gain of 330; about the same as available from the 6AU6 pentode valve (Mullard Reference Manual of Transistor Circuits, 1960). from output back to input). Field-effect transistors (FETs) JFETs (junction FETs) and Mosfets (metal oxide semiconductor FETs) are important types of transistors and will be covered in some detail in the following article next month. Why are they called “transistors”? Many references state that the name “transistor” is a combination of ‘transFeedback in transistors fer/transconductance’ and resistor. As triodes, junction transistors However, a May 28th, 1946 survey exhibit considerable collector-base conducted by Bell Labs offered “a discapacitance, and this has the same cussion of some proposed names”. The circuit effect as for valves – reduc- list encompassed the awkward (“surtion of input impedance and poten- face state triode”) and the whimsical tial oscillation. (“iotatron”). The first-generation alloyed-­ The successful candidate, “transisjunction transistors suffered particu- tor” was “an abbreviated combination larly from collector-base capacitance. of the words ‘transconductance’ (or OC44/45 specifications show feedback ‘transfer’) and ‘varistor’” – see www. capacitances of some 10pF. Given that beatriceco.com/bti/porticus/bell/pdf/ valve triodes proved unworkable with transistorname.pdf these kinds of capacitances, how were Some other sources differ on just transistor triodes used? how the name was arrived at, but this Transistor base-emitter junctions at least seems credible. are forward-biased. This means that transistors present a very low input Summary impedance. At audio frequencies, the Solid-state devices had been demonOC44 has an input impedance of some strated before the start of the 20th cen2.5kW. This low impedance reduces tury and were well-known by 1920. the effect of collector-base feedback, Julius Lilienfeld patented the two and such feedback has little effect at types of amplifying devices that we audio frequencies. now recognise as transistors in the In RF/IF amplifiers, collector-base mid-1930s. His patents were not comfeedback can be so severe as to reduce mercialised, though the Bell Laboratogain or provoke oscillation. The ries team referenced them, and Lilienunwanted in-circuit feedback is com- feld’s patents did forestall some lines plex due to combined capacitive and of enquiry at Bell Labs. resistive effects. The most thorough Building on the development of designs add external components to microwave diode detectors during provide unilateralisation – a fancy World War II, a Bell Laboratories team, name for “single-direction” (signal including John Bardeen and Walter flow is only from input to output, not Brattain and led by William Shockley, published details of the point-­ contact transistor. The device showed the practicalities of solid-state design, but it was difficult to manufacture, fragile and unusable in some circuit configurations. Slightly in advance of Bardeen and Brattain, Welker and Mataré (working in France) released the Transistron, which was successfully taken up by French telecommunications companies. Fig.42: a Mosfet is similar to a JFET, but instead of using a reverse-biased PN Shockley had not been named on junction to isolate the gate from the channel, it uses an extremely thin layer of the landmark Bell Labs’ patent. His semiconductor oxide. The gate's electric field typically enhances electron/hole inventive spirit drove him to develop flow in the channel when applied; it is pinched off otherwise. These are thus and patent the junction transistor – the known as ‘enhancement mode’ devices. More on Mosfets next month. 48 Silicon Chip Australia's electronics magazine siliconchip.com.au device we now universally recognised as the transistor. But Shockley’s patent was a theoretical paper, showing the principles but not the manufacturing details. The first grown-junction transistor created a single crystal device that exhibited much more stable, predictable and reliable characteristics than point-contact designs. However, it suffered from poor high-frequency operation due to its thick base layer. The alloyed junction design, using much more controllable doping by diffusion at near melting-point temperatures, offered much thinner base layers and could operate to 30MHz. This was improved on by the drift-field design, which employed graded doping across the base and pushed frequency limits to 60MHz. The alloy-diffused design abandoned the two-sided construction of all types so far, building the transistor over the collector substrate. The base was diffused into the collector, followed by emitter alloying into the base layer. The Mesa design further developed the all-diffusion process. The final design – epitaxial planar V – uses a thin, lightly-doped epitaxial layer over a heavily-doped substrate giving low resistance; together, these form the collector. The base and emitter are diffused into the collector substrate. Photolithographic masking allows transistors to be fabricated to tiny sizes, with outstanding reliability and reproducibility. The first transistors were fabricated in germanium. Germanium’s temperature sensitivity, leakage, scarcity and its oxide surface’s solubility led to its replacement by silicon as a feedstock. Although alloyed silicon processing was initially more difficult to engineer, its advantages over germanium have seen germanium phased out. References/links • The most comprehensive and best single collection of references available (created by Mark P. D. Burgess): siliconchip.com.au/link/abcj • A replica and description of the first transistor: siliconchip.com.au/ link/abce • A fine general history of transistors: siliconchip.com.au/link/abcf • A detailed description and analysis by van Zeghbroeck: siliconchip.com. au/link/abcg • On diffused transistors generally: https://w.wiki/4fiz • Early History of Transistors in Germany, Herzog, R., 2001: siliconchip. com.au/link/abch • Transistor Production Techniques Next month at RCA, Fahnestock, J. D. ElectronThis article has described the tran- ics, October 1953: siliconchip.com. sistor mass-production techniques au/link/abci that are still in use today. The follow- • RCA Transistor Manual 1964, Radio ing article will explain in more detail Corporation of America how a transistor works, including • Crystal Fire, Riordan, M., and Hodbipolar transistors as well as field-­ deson, L., W. W. Norton and Company, effect transistors (FETs). It will also ISBN 13:978-0-393-31851-7 give some pertinent performance • History of Semiconductor Engicharacteristics, including a descrip- neering, Lojek, B. Springer-Verlag tion of the limitations of transistor Berlin Heidelberg, ISBN-13 978-3SC performance. 540-34257-1 intage Radio Collection March 1988 – December 2019 Updated with over 30 years of content Includes every Vintage Radio article published in Silicon Chip from March 1988 to December 2019. In total it contains 404 (not an error) articles to read, or nearly 150 more articles than before. Supplied as quality PDFs on a 32GB custom USB All articles are supplied at 300DPI, providing a more detailed image over even the print magazine. Physical and digital versions available Buying the USB gives you access to the downloadable copies at no extra charge. Or if you prefer, you can just buy the download version of the Collection. Own the old collection on DVD? If you already purchased the previous Collection on DVD, you can buy this updated version for the discounted price of $30 on USB (plus postage), or $20 for the download version. $50 PDF Download SC4721 siliconchip.com.au/Shop/3/4721 $70 USB + Download SC6139 siliconchip.com.au/Shop/3/6139 Postage is $10 within Australia for the USB. See our website for overseas & express post rates. siliconchip.com.au Australia's electronics magazine April 2022  49