Semiconductor doping
A Socratic walk-through of semiconductor doping — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does adding one impurity atom per million transform a poor conductor into a useful one?
One phosphorus atom for every million silicon atoms sounds like contamination too slight to matter — a rounding error in the purity spec. Yet it takes silicon from a resistivity of roughly two hundred thousand ohm-centimetres down to about a tenth of one, six orders of magnitude, and turns an inert crystal into the material every processor is built from.
The usual explanation is that the impurity brings extra electrons. That cannot be the whole story, because silicon is not short of electrons. A cubic centimetre holds about five times ten to the twenty-second silicon atoms, each contributing four valence electrons, all of them doing nothing useful. So the shortage is not of electrons. What is it a shortage of?
Reasoning it through
REASONING #Ask what a current actually requires. Not electrons, but electrons that can change their state of motion — an electron accelerated by a field has to end up somewhere it was not, and if every nearby state of slightly higher energy is occupied, it cannot go. In pure silicon every bonding state is filled and the nearest empty ones sit 1.12 electronvolts above, across a gap with no states in it at all. The electrons are there; the vacancies are not.
How many cross on their own? Thermal energy at room temperature is about 0.026 electronvolts, forty times smaller than the gap, so crossing is exponentially unlikely — the mobile carrier density in pure silicon works out to roughly ten to the tenth per cubic centimetre. Hold those two numbers together: five times ten to the twenty-second atoms, ten to the tenth carriers. About one atom in five trillion is contributing anything.
Now the "one in a million" looks entirely different. It puts five times ten to the sixteenth phosphorus atoms into that cubic centimetre. Compare it not to the atom count, which was never the relevant baseline, but to the carrier count it competes with — and it is five million times larger. The intuition that a part per million must be a small perturbation used the wrong denominator.
But why should a phosphorus atom donate anything? It sits in a silicon site and uses four of its five valence electrons for bonds, leaving the fifth attached to a core that now carries one extra positive charge. That is a hydrogen atom in miniature — except that the attraction is screened by silicon's dielectric constant of about 11.7, and the electron moves with an effective mass around a quarter of a free electron's. Scale hydrogen's 13.6 electronvolts by the effective mass and by the square of the dielectric constant and you get something near 0.03 electronvolts. The measured value for phosphorus in silicon is 0.045 — the right order.
That number is the whole trick. The donor state sits a few hundredths of an electronvolt below the empty band, not 1.12 above the full one. Thermal energy of 0.026 electronvolts cannot lift an electron across the gap, but it easily lifts one off a donor. So the gap stays impassable while the impurity is essentially fully ionised at room temperature. Boron does the mirror image: one bond short, an empty state just above the full band, thermal energy promotes an electron into it, and the mobile vacancy left behind carries current the other way.
The analogy
THE ANALOGY #Picture a cinema at full capacity with an empty balcony forty rows up a sealed staircase. Nobody moves, not because the room is empty but because there is nowhere to move to and the balcony is out of reach. Doping does not add people. It puts a handful of chairs on a landing one step off the floor — and one step is a height anyone can manage. A few dozen relocate, and now the whole floor can shuffle.
Cinema-goers choose to move and the seats are countable, whereas occupancy here is set by a temperature-dependent statistical distribution — cool the crystal and the donors stop ionising, as if the people froze back into their chairs.
Clarifying the model
THE MODEL #Three refinements, because the tidy story oversells itself in three places.
First, more doping does not mean indefinitely more conduction. Each ionised dopant is a charged scattering centre, so mobility falls as doping rises — from around 1400 square centimetres per volt-second in lightly doped silicon to a few hundred at high levels. Conductivity is carrier density times mobility, and the two work against each other, so device doping is a compromise, not a maximum.
Second, "impurity" is not the operative property; shallowness is. Gold in silicon is also an impurity, sits near the middle of the gap, and does the opposite of doping — rather than supplying carriers it traps and recombines them, and was historically added on purpose to make fast switching diodes. Same picture, a level in a different place, an opposite effect.
Third, the hydrogen-like calculation is an approximation that works because the donor electron's orbit spans several nanometres, many lattice spacings, so it genuinely sees an averaged medium. It fails for deep levels, and it fails at the top end: above a few times ten to the eighteenth per cubic centimetre the donor orbits overlap, the levels smear into a band merging with the conduction band, and the material stops behaving as a semiconductor at all. Exactly where that transition sits is still discussed.
And the deepest point is easy to miss: conductivity was never the goal. Copper conducts better than any doped silicon. What doping buys is conduction whose magnitude, sign, and location are chosen — n-type here, p-type a micrometre away, and a junction between them that passes current one way and can be switched by a field. The useful property is control, not conduction.
A picture of it
THE PICTURE #How to readHeight on the page is electron energy, increasing upwards, and each band is a range of states rather than a place. The bottom band is full and the top one empty, which is why pure silicon conducts so badly: electrons and vacancies are separated by the middle strip, whose label gives the comparison that decides everything — 1.12 electronvolts of gap against 0.026 of thermal energy. The two small blocks are what doping adds. They are not in the bands; they sit in the gap, a couple of hundredths of an electronvolt from an edge, close enough that room temperature empties the upper one into the band above and fills the lower one from the band below.
What became clearer
WHAT CLEARED #The scarce thing in silicon was never electrons — it was accessible empty states, and the 1.12 electronvolt gap puts them out of thermal reach. A part-per-million dopant looks negligible only against the atom count; against the ten-to-the-tenth carriers actually present it is a factor of a million. And it works because the level it introduces sits just below or above a band edge, so ordinary thermal energy ionises it while leaving the main gap untouched. Shallowness, not impurity, is the property that matters — which is why a differently placed impurity kills carriers instead of supplying them.
Where to go next
ONWARD #- What happens at the boundary where n-type meets p-type, and why that junction rectifies.
- Why doped silicon becomes less conductive as it is heated, while pure silicon becomes more so.
Key terms
TERMS #| Term | What it means |
|---|---|
| Intrinsic carrier concentration | the density of mobile electrons and holes in an undoped semiconductor, about ten to the tenth per cubic centimetre for silicon at room temperature. |
| Shallow level | an impurity state lying only a few hundredths of an electronvolt from a band edge, and therefore ionised at room temperature. |
| Mobility | how readily a carrier drifts under an applied field, reduced by scattering from the very dopants that supplied it. |
Every term the collection defines is gathered in the glossary.