The MOSFET

A MOSFET looks impossible at first inspection, and understanding why it is not is the whole physics of the device. An n-channel MOSFET is built on P-TYPE silicon, whose mobile carriers are holes. Into it are diffused two heavily doped N-TYPE regions, the source and the drain. Between them is p-type silicon, and above that a thin oxide with a metal gate on top. With no gate voltage, look at the path from source to drain: n-type, then p-type, then n-type. That is two junctions back to back, one of which is reverse-biased whichever way you apply a voltage. No current flows. The device is off, and it is off very well. Now put a positive voltage on the gate. The gate, oxide and silicon form a capacitor, so positive charge on the gate must be balanced by negative charge in the silicon. At first that happens by pushing the mobile holes away from the surface, leaving behind the fixed negative charge of the acceptor atoms — depletion. Raise the gate voltage further and the depletion region cannot supply any more negative charge, because the acceptors are fixed in place and there are only so many. So the silicon does something else. It pulls ELECTRONS to the surface — minority carriers, thermally generated and supplied readily by the two n-type regions sitting right there. At a certain gate voltage the electron concentration at the surface exceeds the hole concentration, and a thin sheet of the p-type silicon has become effectively n-type. It has INVERTED. That inversion layer is a continuous n-type path from source to drain, and current flows. The gate voltage at which it forms is the THRESHOLD VOLTAGE, and it is the single most important parameter of the device. Three consequences worth holding onto. The channel is not built, it is INDUCED. Remove the gate voltage and it vanishes. Nothing was doped, nothing was etched; the conducting path is a population of electrons held at the surface by an electric field. The gate draws no steady current, because there is an insulator under it. It draws a transient current while the capacitance charges, and that is all. This is the property that lets you put billions of them on a chip. And the source and drain are physically identical. Which one is which is decided by the voltages you apply, not by the construction — the device is symmetric, and in some circuits current genuinely flows both ways.

WHY DID LILIENFELD'S 1926 PATENT NOT WORK? Surface states. He had the physics right and no way to make a surface clean enough. Bardeen's 1947 surface-state theory explained the failure quantitatively: trapped charge at the semiconductor surface screens the applied field so thoroughly that almost none penetrates. Attempting to verify that theory is what led Bardeen and Brattain to the point-contact transistor — the bipolar transistor exists because the field-effect transistor did not. WHY DID THE THRESHOLD DRIFT IN EARLY DEVICES? Sodium. Sodium ions are small, positively charged and mobile in silicon dioxide at modest temperatures, so they drift under the gate field and pile up at whichever interface the field pushes them to. Positive charge near the silicon shifts the threshold, so a device biased on for hours came back with a different threshold. It was traced to contamination from handling and glassware, and defeated by obsessive cleanliness plus phosphosilicate glass layers that trap the sodium before it arrives. Almost the entire discipline of semiconductor cleanroom practice descends from chasing this one ion. WHAT WAS THE SILICON GATE, AND WHY DID IT MATTER SO MUCH? Original MOSFETs used an aluminium gate, deposited AFTER the source and drain were diffused — so the gate had to be aligned to them, and alignment error meant you had to make the gate deliberately oversized to guarantee overlap. That overlap is capacitance between gate and source or drain, and it slows the device badly. The silicon-gate process, developed by Federico Faggin and Tom Klein at Fairchild around 1968 building on work by Chih-Tang Sah, replaces the metal with heavily doped polycrystalline silicon and deposits it BEFORE the source and drain. The gate then acts as its own mask: the dopant enters everywhere except where the gate shadows it, so the source and drain define themselves against the gate edge with zero alignment error. That is SELF-ALIGNMENT, it removes the overlap capacitance almost entirely, and it is the single change that made MOS fast enough to build a processor from. Faggin took it to Intel and used it to design the 4004. WHY IS CMOS THE ONE THAT SURVIVED? A single MOSFET switching a resistive load burns power whenever it is on. CMOS pairs an n-channel and a p-channel device so that in either stable state exactly one of them is off — so the path from supply to ground is always broken and static current is essentially zero. Frank Wanlass patented it at Fairchild in 1963. It was ignored for years because it needs both device types and therefore more process steps and more area. It won when power density, not area, became the binding constraint, which happened in the 1980s and has never stopped being true. WHY DID THE GATE OXIDE STOP SHRINKING? It reached about 1.2 nanometres, five atomic layers, and electrons began tunnelling straight through it. Leakage through the gate started to rival the switching power. The fix was to stop using silicon's own oxide for the gate and adopt hafnium-based high-permittivity dielectrics around 2007, which give the same capacitance from a physically thicker film. Fifty years of running on the material's own oxide ended there, for a quantum-mechanical reason. WHAT IS THE HONEST SUMMARY? The MOSFET was invented first, worked last, was unimpressive when it finally worked, and then turned out to have the one property — no static control current — that mattered more than everything the bipolar transistor was better at. That is a reasonably common shape for a technology, and it is worth remembering the next time something is dismissed for being slower.