Switch and amplifier: BJT vs MOSFET

Last time, Bell Labs built a little crystal device that let a small signal control a big one . That sounds like getting something for nothing . A whisper goes in, a shout comes out. So where does the extra come from? And how can a few layers of doped silicon steer a current they don't supply? There are two answers, two kinds of transistor, and the one that lost the first round ended up in almost everything you own . First, nothing makes energy out of nothing . An amplifier is more like power steering . Your hands supply a small effort, but most of the force that swings the wheels comes from a pump driven by the engine . Your hands mostly just steer it . Switch the engine off and the steering goes heavy . A transistor amplifier works the same way. The big current comes from a power supply, like a battery, and the transistor is a valve that shapes that flow into a bigger copy of the small signal . So "gain" is just a ratio: how big the output is, divided by how big the input is . If ten millionths of an amp going in controls one thousandth of an amp coming out, that's a current gain of a hundred . A hundred is a typical figure for a good transistor of the first kind . Let's open one up. The first kind is the bipolar junction transistor, or BJT . It's a sandwich of doped silicon: an n-type layer called the emitter, a very thin p-type layer called the base, and another n-type layer called the collector . That's two of the junctions from episode three, back to back . Put a voltage across the whole thing and, on its own, almost nothing flows, because one of the two junctions is always blocking . That's "off" . Now give the base a small push: about six tenths of a volt between base and emitter . That opens the first junction, and the emitter floods the base with electrons . But the base is p-type. Electrons there are the outsiders, the minority carriers from the doping episode . In p-type silicon they don't last long. Sooner or later, each one meets a hole and disappears . Unless the base is thin. In a modern transistor, the base is about a tenth of a micron wide, while an electron could typically wander tens of microns before it met a hole . So almost every electron makes it across . On the far side, the collector junction is blocking, but blocking only stops carriers coming the other way . Its electric field grabs every electron that arrives and sweeps it into the collector . About ninety-nine per cent of the current leaving the emitter arrives at the collector . So what does the base current actually pay for ? Mostly, it's holes leaking backwards from the base into the emitter, a side effect of opening that first junction . The trick to keeping it small is doping . Make the emitter far more heavily doped than the base, and the emitter sends across a hundred or so electrons for every hole it gets back . That ratio is the current gain . And here's the part that makes amplification so powerful . The current through a BJT depends on the base voltage exponentially . Raise the base voltage by just sixty millivolts and the current goes up about ten times . So in a simple circuit, a wiggle of a single millivolt at the input can become a swing of about a fifth of a volt at the output, with the energy coming from the supply . The same device can be a switch or an amplifier, depending on how hard you drive it . With no base current, it's off: cut-off . Push the base hard and it's fully on, with only about two tenths of a volt left across it: that's called saturation . In between is the active region, where the collector current follows the base current in proportion . An amplifier keeps the valve half open and wiggles it. A switch slams it shut or throws it wide. Computers use the ends. Radios and hi-fi use the middle . But the BJT has a catch . To stay on, it needs a steady base current flowing in, all the time . That's fine for one transistor. On a chip with thousands of them, the currents that keep them all switched on add up to heat and wasted power . And on a chip, each bipolar transistor needs extra structure, an isolation moat, to stop it interfering with its neighbours . Engineers had dreamed of a better way since before the BJT existed . In the nineteen-twenties, Julius Lilienfeld filed patents on the idea of controlling the current in a semiconductor with an electric field . There's no evidence he ever built one that worked . Bell Labs tried too, before and after the war . As we saw in the 1947 story, charges stuck at the surface blocked the field, and that search led to the point-contact transistor instead . The fix came from silicon's skin of glass, the oxide from the first episode . At Bell Labs, John Atalla's group found that growing silicon dioxide on the surface greatly reduced those troublesome surface states . With far fewer of them in the way, an electric field could finally reach into the silicon . By the end of 1959, Atalla and Dawon Kahng had made the first insulated-gate field-effect transistor that worked . They announced it in 1960 . Bell Labs didn't pursue it . It was slow, and the telephone system had no pressing need for it . But in a 1961 memo, Kahng pointed to its ease of fabrication, and its possible use in integrated circuits . Its name describes the layers . A metal gate, on a thin oxide, on a semiconductor: MOS . FET means field-effect transistor . Underneath is p-type silicon, with two n-type islands set into it, called the source and the drain . With the gate at zero volts, going from source to drain means crossing two junctions, and one of them always blocks . Off. Now put a positive voltage on the gate . Its electric field reaches through the glass and pushes the holes away from the surface . Push harder, past a threshold, typically around four tenths of a volt, and it draws electrons in from the source and drain, into a sheet right under the glass . That thin layer, only a few nanometres thick, has flipped from p-type to n-type, so it's called an inversion layer . It joins source to drain, and current flows . The gate never touches the channel . It summons one. And because the gate sits on an insulator, no steady current flows into it . A MOSFET's input resistance can be a trillion ohms or more . So, unlike a BJT, its gate can hold it on or off while drawing almost nothing . It's not quite free. The gate behaves like a tiny capacitor, so every time it switches, a little charge has to flow in or out . But the gate itself only costs energy when it changes, not to stand still . For years, few people expected much of it . An RCA research head said it was slow and would never threaten bipolar transistors . The first commercial MOS chip arrived in 1964 . But early devices drifted . Gordon Moore said it was easy to make one that worked, but they couldn't make one whose behaviour stayed stable . One culprit, found at Fairchild in the mid-sixties, was sodium contamination in the oxide . By the early nineteen-seventies, the problems were solved . Then MOS won the chip . A 1970 Philips handbook gave the reasons plainly . Most important, MOS circuits need no isolation diffusions . They take fewer steps to make, with fewer masks . And each transistor can be smaller . So far more logic fits on a chip the same size . Bipolar was still faster: back then, MOS logic topped out at a few megahertz, and anything much faster than ten megahertz needed bipolar . The real winner was scaling . In 1974, Robert Dennard's group at IBM showed that if you shrink a MOSFET and lower its voltage in step, it gets faster and uses less power, while the power per square millimetre stays about the same . Each new generation shrank features to about seventy per cent, halving the area . The gate oxide went from three hundred nanometres to one point two, about five atomic layers . By one estimate, by 2018 the industry had shipped thirteen sextillion transistors, and ninety-nine point nine per cent of them were MOS . So is the BJT finished ? Not quite. Its high speed, low noise and output power still make it the choice in some high-frequency and analogue circuits, like parts of a phone's radio . Often, a few BJTs sit on a chip full of MOSFETs . And in power electronics, around 1980, Jayant Baliga at General Electric combined the two: a MOS gate controlling a bipolar current . That device, the IGBT, now switches power in cars, appliances, solar panels and trains . So now we have a switch whose gate holds it on or off for almost no power, and which, for decades, got better the smaller you made it . A modern chip has billions of them . Nobody wires those up one at a time. So how do you make billions of transistors, side by side, all at once, on one slice of silicon? You print them, with light. That's next time.