The Hall Effect

Intuition

Reading a current without touching the wire

A clamp meter that measures alternating current can be a coil and nothing else, because an alternating current makes a changing field and a changing field makes a voltage in a coil. Ask that same meter to read the direct current from a battery and it goes blank. The field is there, steady and perfectly measurable, but a coil only responds to change, and nothing is changing.

The instrument that fills that gap does something different. It puts a thin sliver of semiconductor into the field, passes a small current along it, and reads a voltage that appears across the sliver, at right angles to both. That voltage is proportional to the field and it does not care whether the field is steady. Edwin Hall found the effect in 1879, and for a century it was mostly a laboratory curiosity for measuring what materials are made of.

What changed is that the same sliver turned out to be an ideal magnetic sensor. It has no moving parts, it works down to zero frequency, it costs very little, and it can be built on the same silicon as the amplifier that reads it. Hall devices now sit inside DC clamp meters, motor controllers that need to know where the rotor is, contactless switches, and the current-sensing loop of most things that charge a battery.

Practitioner

The sideways voltage, and what sets its size

Drive a current along a flat plate and put a field through the plate's face. The moving carriers are pushed sideways, they pile up along one edge, and the charge separation builds a voltage — until that voltage pushes back hard enough to cancel the sideways force. The plate settles there, and the voltage it settles at is the Hall voltage.

The Hall coefficient R_H is the material's contribution, and thickness appears in the denominator because a thinner plate crowds the same current into a smaller cross-section.

Three properties of that relationship do the practical work. It is linear in the field, so a Hall device makes a well-behaved analogue sensor rather than a threshold detector. It is linear in the drive current, which means the output is proportional to the product of two inputs — the basis of Hall-effect multipliers and wattmeters. And it reverses sign with the field, so a Hall sensor distinguishes a north pole from a south pole, which a coil-based sensor at rest cannot do at all.

Twelve millivolts from a hundred millitesla is small but entirely workable, and it is the reason a modern Hall sensor arrives as a three-terminal device with the amplifier already inside it rather than as a bare plate.

The drive current is a design variable rather than a constant of nature, and raising it raises the output in proportion — up to the point where the plate's own dissipation starts heating it, which shifts both the sensitivity and the offset. Integrated devices settle that trade internally and present you with a supply pin instead, which is why a datasheet quotes sensitivity in volts per tesla rather than the raw Hall coefficient. Where you do meet a bare element, in a laboratory gaussmeter probe, the drive current is specified alongside the calibration and changing it invalidates both.

One point of vocabulary saves confusion later. What the plate responds to is flux density at its own face, in tesla, not the total flux through some larger area. A Hall device is a point sensor, and putting it in a wider gap does not average the field across that gap; it reads whatever is happening in the few square millimetres it occupies, which is why the mechanical placement of a current sensor in its core matters as much as the core does.

Engineer

Why the plate has to be a semiconductor

The Hall coefficient is not an arbitrary material property. It is the reciprocal of the carrier density multiplied by the carrier charge, so a material with few carriers has a large coefficient, and one with many has a small one. Metals conduct well precisely because they have enormous carrier densities, and that is exactly what ruins them as Hall elements.

That figure is the free-electron-model value, the textbook estimate for a metal with one conduction electron per atom. The measured coefficient is closer to −5.3×10⁻¹¹ m³/C — smaller in magnitude, and negative, the sign a direct signature of electrons rather than holes doing the conducting. Either number makes the same point: copper's coefficient is minuscule next to doped silicon's.

A nanovolt and a half is below the noise of any practical amplifier and drifts with temperature by more than its own value. Doped silicon and the compound semiconductors used for the best devices have carrier densities millions of times lower, and they buy their sensitivity with exactly that scarcity. This is the same trade a semiconductor's doping level always represents: a controllable, deliberately small population of carriers is worth far more than a large uncontrollable one.

The sign of the Hall voltage carries information too. It tells you whether the carriers doing the conducting are negative or positive, and in a p-type semiconductor the answer really is positive — holes. That measurement was one of the earliest hard confirmations that hole conduction is not merely a bookkeeping convention, and it is still how carrier type and concentration are measured in a fabrication line.

Geometry limits the model in a way the expression does not admit. It assumes a plate much wider than it is thick and much longer than it is wide, so that the current is uniform and the drive contacts sit far from the sensing ones; a squat element short-circuits part of the Hall voltage back through its own drive contacts and gives measurably less than the relationship predicts. Manufacturers quote a geometric correction factor for exactly this reason, and it is one of the things you are paying for when you buy a device rather than etching a plate.

Orientation limits it too. The relationship counts only the field component perpendicular to the plate's face, so a tilted field contributes its cosine and nothing more — the same projection magnetic flux sets out, and the same trap, because a sensor mounted at sixty degrees reads exactly half and gives no sign that anything is amiss.

Professional

What a Hall sensor is really like to design with

Hall devices arrive in three broad forms, and picking the wrong one costs a redesign. A switch has a comparator built in and gives a clean digital output above a threshold field, which suits end-stop and lid detection. A latch switches on one polarity and stays put until it sees the other, which is what brushless motor commutation wants. A linear device gives an analogue output proportional to the field, which is what current sensing needs.

Current sensing is the application worth working through, because the whole chain is multiplication and every stage costs accuracy.

Now the awkward part. That output sits on top of the device's own offset, which is a voltage present with no field at all, and offset refers back through the same chain into an apparent current.

Two amps of error is invisible at 30 A and useless at 2 A, which is why Hall current sensors are specified by their full-scale range and are a poor choice for measuring small currents in a wide-range design. It is also why serious devices use chopper stabilisation, rotating the drive and sense contacts electrically at kilohertz rates so that the offset averages out while the true Hall signal does not. That technique is the single biggest reason modern Hall current sensors are usable at all, and it is also the reason they inject a small switching artefact into their own output.

Temperature is the other standing problem. Carrier density in a doped semiconductor changes with temperature, so both sensitivity and offset drift, and a datasheet's accuracy figure is nearly always dominated by temperature rather than by anything you can fix in the circuit. Ratiometric devices, whose output scales with their supply, cancel one term of this when the ADC shares the same reference — a useful trick that works only if you actually share the reference.

The alternatives are worth knowing. A shunt resistor is cheaper and more accurate but is galvanically connected to the circuit and dissipates power. A current transformer is isolated, accurate and lossless but cannot read DC, for the reason Layer 1 opened with. Magnetoresistive sensors beat Hall devices on sensitivity and drift and cost more. Hall wins where you need isolation, DC response and low cost together, which is a common enough corner to have made it ubiquitous.