Yet Another Semiconductor V/I Curve Tracer – I

The measurement, before the circuit

A curve tracer plots the current through a device against the voltage across it; the ramp, the step ladder and the current limit exist only to move that point around the plane in a way the device survives. The scope does the plotting: in X-Y mode the timebase is disconnected, channel 1 drives the beam horizontally and channel 2 vertically.

That channel must carry the voltage at the device terminals, not at the driving source. The two differ by the drop across the sense and limiting resistors; plot the source and every ohm in the path becomes a slope error through the origin.

A bipolar ramp from a DAC

A sweep has to cross zero, because the interesting half of a diode is behind the origin. The MCP4822 is a dual 12-bit DAC with an internal 2.048 V reference, a selectable 2x buffer, ±2 LSB typical INL and 4.5 µs settling: 0 to 4.096 V in 1 mV steps, which an op-amp stage with gain and offset turns into ±10 V.

That 1 mV step hides the problem. Twelve bits across ±10 V is 4.88 mV per code, while a junction moves about 60 mV per decade: one LSB jumps the current by a fifth, so the knee is a staircase of twelve steps per decade. Two ranges — ±1 V near the knee, ±30 V for breakdown — buy the resolution back.

An ATmega32 has no DAC, only four PWM channels and a 16-bit Timer1; filtered 8-bit fast PWM at 62.5 kHz smooths easily, but 8 bits over 20 V is 78 mV per step.

Borrowing a ramp from the mains

A transformer secondary is the other classic ramp: isolated, 50 or 60 Hz, no code. Drawn both ways, a capacitor shows its hysteresis loop and the brightness of the return path hints at heating within the cycle.

Two costs. A sine dwells near its peaks, where dv/dt approaches zero, so the device sits at maximum dissipation for much of every cycle — the opposite of what a fragile part wants. And the amplitude is set by the winding, so range switching means switching taps.

The isolation comes from the transformer; a Variac is not an isolation device. The scope’s ground clip is bonded to protective earth, so clipping it to a live node closes a circuit through the mains wiring.

Current limiting and the sense resistor

A resistor between source and device sets the peak current, roughly (V_peak − V_device)/R_s. Ten volts across 100 Ω is 100 mA, and that resistor dissipates a watt at the peak, continuously. The device gets 0.5 W — about the whole rating of a small TO-92 at 25 °C with no heatsink. The curve that shows second breakdown is drawn once.

The sense resistor sets the vertical scale. One ohm in the return leg gives 1 V/A, so 100 mV/div is 100 mA, and the 1 MΩ scope input shunts it into insignificance. At 100 kΩ that input pulls the reading down by 9% — a calibration term, and the range leakage needs.

Both channels share one ground, so the sense resistor belongs in the low side and the device’s common terminal at earth; a grounded cathode needs a differential probe.

Stepping for a family of curves

The second DAC channel drives the base through R_b, and each rung draws one trace. Since I_b = (V_step − V_be)/R_b and V_be moves only about 0.1 V between 10 µA and 10 mA, a 100 kΩ base resistor turns a 10 V step into 100 µA with about 1% error; at 1 kΩ the family is visibly uneven.

Eight to ten steps fill a screen without overlap. The step must be static while the sweep runs, so update it at the zero crossing and let the DAC settle. A MOSFET is easier: negligible gate current, linear spacing by construction.

Reading the trace: knee, breakdown, leakage

The knee is not a point: a junction conducts over roughly 0.5 V to 0.7 V and the transition is exponential, so its apparent position depends on vertical gain. On a log current axis the forward characteristic is a straight line of slope 2.3·n·V_T; at 300 K that is 59.5 mV per decade at n = 1.

Breakdown lives at the far left. The 1N4148 specifies 100 V minimum breakdown at 100 µA but only 75 V of continuous reverse standoff: the breakdown figure is a limit, not an operating point. A zener is built to sit there: below roughly 5 V to 6 V breakdown is Zener tunnelling with a negative temperature coefficient, above it avalanche with a positive one, and near 5.6 V the two cancel — which is why a 5.6 V zener traces a sharp corner and a 3.3 V one creeps as the die warms.

Leakage is where the vertical scale betrays you. The 1N4148 specifies 25 nA at 20 V, 25 °C, and 50 µA at 150 °C: 2000× over 125 °C, a doubling every 11 °C. At 1 mA/div, 25 nA is a fortieth of a division.

Every number above is a datasheet figure or arithmetic on one; nothing was built or measured. A builder must still check the output range of their DAC stage, the settling of their op-amp at the sweep rate, and whether the scope’s vertical amplifier is DC-coupled at the chosen deflection factor — an AC-coupled channel turns a horizontal line into a banana.

References

• Microchip, MCP4802/4812/4822 DAC Data Sheet, DS20002249.
• Microchip (Atmel), ATmega32/L Data Sheet, doc2503.
• Vishay, 1N4148 Data Sheet, reverse-current table.
• Toshiba, Zener diode temperature coefficient, technical FAQ.