The Planar Process

Hoerni's fix only makes sense if you understand what it was fixing, and surface leakage is invisible until you measure it. WHAT GOES WRONG AT A BARE SURFACE. Inside the crystal every silicon atom has four neighbours and four satisfied bonds. At a surface the lattice simply stops, leaving unsatisfied bonds — dangling bonds — that sit at energies inside the forbidden gap and can trap or release charge. A trapped charge is an electric field, and an electric field at the surface of a junction bends the bands and creates a conducting channel along the surface where none was designed. Current leaks around the junction rather than through it. Worse, the surface charge is not fixed. Water vapour adsorbs and desorbs. Sodium ions — from fingerprints, from glassware, from the air — are small and mobile and drift under any applied field, so the leakage changes as the device is used and changes again when it is switched off. This is why early transistors had reputations rather than specifications. WHAT THE OXIDE DOES. A thermally grown oxide satisfies most of those dangling bonds by bonding to them. The residual interface state density on a well-grown, hydrogen-annealed silicon-silicon dioxide interface is around 10^10 per square centimetre, against roughly 10^15 atoms per square centimetre at the surface — so fewer than one site in a hundred thousand remains electrically troublesome. Nothing else in materials science comes close, and it is specific to this one pairing. SEE IT FOR YOURSELF, WITH A DIODE. You cannot fabricate a junction, but you can demonstrate that surface condition changes leakage, on any ordinary silicon diode with a glass body. Set up a reverse-bias leakage measurement: diode reverse-biased through a large resistor, say 10 megohms, from a 9 V supply, with your multimeter measuring the voltage across the resistor. Reverse leakage in a small signal diode is nanoamps, so across 10 megohms that is tens of millivolts — measurable. Now change the surface and watch the reading. Breathe on the glass body: the reading moves as moisture condenses, then drifts back as it evaporates. Handle the body with bare fingers and measure again; the salt and oil from a fingerprint is a conductive film and an ion source. Warm it gently with a hairdryer and watch leakage climb — some of that is genuine thermal generation, which doubles roughly every 8 to 10 degrees, but the surface contribution is what recovers slowly rather than instantly when you stop heating. What you are watching is a device whose junction is sealed inside glass, and it STILL responds to what happens on its outside surface. Now imagine the junction itself exposed. That is the mesa transistor, and that is what Hoerni buried.

WAS THE PLANAR PROCESS OBVIOUS? It looks obvious now, which is the usual fate of a good idea. At the time the oxide was universally regarded as a nuisance to be removed once it had served as a diffusion mask, partly because nobody was certain it was not itself a source of contamination. Hoerni's contribution was to recognise that the layer everyone was throwing away was the solution to the problem everyone was fighting. His notebook entry is dated 1 December 1957; he did not build one until 1958 because Fairchild was busy shipping mesa devices. WHY DID EPITAXY BECOME PART OF IT? A transistor wants a lightly doped collector for high breakdown voltage but a heavily doped path to the contact for low resistance, and one uniformly doped wafer cannot be both. Epitaxy — growing a fresh, lightly doped single-crystal layer on top of a heavily doped substrate — gives you both in one wafer, with the layer taking its crystal orientation from the substrate beneath. It arrived in 1960, a year after planar, and the two together are what people usually mean by the planar process today. WHAT IS THE REAL COST OF JUNCTION ISOLATION? Area, capacitance and leakage. Every isolated island needs a diffused wall around it, and that wall must be wide enough to allow for lateral diffusion, so a large fraction of early IC area was isolation rather than device. The isolation junction also has capacitance to the substrate, which slows the circuit, and it leaks a little, which matters in low-power designs. It was eventually replaced by dielectric isolation — oxide-filled trenches — which costs process steps instead of area. WHY ALUMINIUM FOR THE METAL? It adheres well to silicon dioxide, it can be evaporated easily, it patterns cleanly with photolithography, and crucially it makes a good ohmic contact to silicon after a modest anneal. Its problems arrived later at scale: aluminium dissolves silicon at the contact, which can spike through a shallow junction, and at high current density the electron wind physically moves aluminium atoms along the wire until it thins and fails — electromigration. Both were managed with alloying and barrier layers before copper replaced aluminium in the late 1990s. WHY IS THIS A PROCESS BLUEPRINT AND NOT A DEVICE ONE? Because the device is ordinary and the process is the invention. A planar transistor does exactly what a mesa transistor does, only reliably. What Hoerni actually created was a manufacturing method with three properties nobody had had together: junctions sealed at birth, a flat surface, and complete compatibility with photolithographic patterning. The next two blueprints are what other people did with those three properties within eighteen months.