Skip to content

Limitations

What the driver does not do, what it assumes where the datasheet is silent, and what no test here can confirm. Every assumption has a knob in Config and a model knob in the tests, and every open question has an owner decision or a bench step. The driver is not verified on hardware: everything below is tested against a model written from the datasheet, never against a real sensor.

Datasheet sections this page relies on

From Microchip Technology Inc. DS20001942 L:

  • Accuracy of each device: DC Electrical Characteristics, Sensor Accuracy, p. 3
  • Thermal response: DC Electrical Characteristics (t_RES), p. 4
  • Thermal considerations and self-heating: 4.4 Thermal Considerations, p. 11

All of them, with the driver API for each: Datasheet map.

Assumptions, each with a knob

ID Item What the driver does Config knob Model knob and test
A-1 The turn-on time has a typical value (800 µs) and no maximum (p. 4) Waits 2400 µs, three times the typical, always power_on_settle_us turn_on_us; test_timing
A-2 The ADC's sample-and-hold settling after a channel switch is your hardware's Discards one conversion before each reading discard_conversions hold_ratio; test_faults
A-3 The noise of your board is not in the datasheet Rejects a spread above 30 mV (3 °C on a 10 mV/°C part) max_spread_mv noise_mv; test_boundaries
A-4 The accuracy is -4/+6 °C worst case (p. 3), and an ADC adds error Accepts 8 °C beyond the specified range on each side range_margin_mc test_boundaries
A-5 The output of an unpowered or switched-off sensor is not in the datasheet Never relies on it; start-up checks the first reading none unpowered_mv; bench step in Troubleshooting
A-6 The thermal response is 1.3 s to 63 % (p. 4), so a long average is stale Caps one reading at one second (a design decision) samples, sample_interval_us kMaxAveragingWindowUs; test_validation
A-7 The ADC's full scale and resolution are yours; a wrong full scale hides a clip Trusts the declaration; documents the effect adc_full_scale_mv, adc_resolution_bits full_scale_mv; test_faults
A-8 Line regulation is 0.1 °C/V and self-heating 0.179 °C; the accuracy is tested at 3.3 V (MCP9700 family) or 5.0 V (MCP9701 family) Does not correct them none use_line_regulation; test_accuracy
A-9 The nonlinearity is ±0.5 °C typical over 0 to +70 °C Leaves it in the result; AN1001 compensation is out of scope none nonlinearity_c; test_model
A-10 The High-option accuracy rows cover -40 to +150 °C and the datasheet does not say an H part also meets the Extended rows High uses the High rows alone over the whole range (the looser, safe reading); closed: no tighter guarantee is reported none needed test_accuracy
A-11 A VOUT shorted to ground reads near 0 V plus the ADC's zero error Accepts it when the ADC reports at or above the window floor (20 mV, 49 mV); Configuration range_margin_mc test_boundaries

Open questions

ID Question State
O-1 Keep the single floating-point helper ToCelsius()? Kept: one pure function, the rest is integer
O-2 The output of an unpowered or floating pin Bench step, not verified (Troubleshooting)
O-3 The MCP9700, MCP9700A and MCP9700B (and any Custom) are one transfer function to the driver The declaration is trusted; only AccuracyLimitsFor() differs by grade
O-4 range = High on a part that is not an H part accepts readings between +125 °C and +158 °C the part does not guarantee The declaration is trusted; the table names the parts that have the option
O-5 The ESP32 calibrated ADC ceiling at 12 dB (about 3.1 V) The example declares 3100 mV; not verified on hardware
O-6 Two-point calibration (a gain trim) Not in this version; a future gain_cal_ppm pivoting on 25 °C is an additive change
O-7 The datasheet's note 3 (p. 3) cites "Equation 4-2" for the first-order equation A datasheet typo: Equation 4-1 (p. 10) is the transfer function, 4-2 is self-heating; the driver follows 4-1
O-8 The output load regulation cell lost its unit in text extraction Read from the PDF: 2 Ω (p. 4); not used by the code
O-9 The TO-92 body is drawn in inches, and its height row is named "bottom to package flat" The millimetre values on Hardware are conversions
O-10 One repository for the whole family Decided: the datasheet lists exactly five devices, held as table rows
O-11 Custom is a declaration, not a verified part Named variants are the default; Custom must declare everything and shares nothing with them
O-12 A Custom whose output is 0 mV at its minimum cannot catch a grounded line Documented below; a test asserts it so it is not mistaken for a defect
O-13 Custom rejects a negative slope and a negative output at its minimum Documented; an additive change if a part needs it
O-14 A floating ADC input that holds a stable mid-scale level looks like a temperature Documented under "What it does not do"; remedy is a board pull-down or a known-temperature start-up check; bench item

What it does not do

  • No correction for nonlinearity, supply or self-heating. The result is the first-order equation of the datasheet; the error budget is on Transfer function. For better accuracy than ±0.5 °C typical after calibration, compensate with a table (the datasheet points to AN1001).
  • No gain trim. One point calibrates the offset only.
  • No detection of a floating pin that holds a stable level. A pin with no sensor that sits at a steady mid-scale voltage (a leaking ADC input, crosstalk, a held sample-and-hold capacitor) passes every check: 1650 mV reads as 115 °C on an MCP9700. A stuck-at-0 line (OutputLow), a rail-high line (Saturated) and a moving line (Unstable) are caught. On the board, add a 100 kΩ pull-down at the ADC pin so an open line reads OutputLow, or compare the first reading with a known temperature (O-14). The range window is a property of the part, so it cannot be narrowed to the climate of one product without a knob; range_margin_mc shrinks it.
  • A grounded line is caught only below the window floor (20 mV or 49 mV by default, A-11).
  • No detection of the part. The five devices cannot be told apart at run time; the declaration is yours.
  • No detection of an ADC that is not what you declared. A full scale that is too high hides a clip.
  • No power measurement. The operating current (6 µA typical) is a documented value, not a reading.
  • No retries. A reading that fails is reported once; call again if you want another.
  • Not thread-safe. One owner, or one mutex.

Custom parts

Custom serves another linear active thermistor IC whose output is first order. It is a declaration, not a verified part: no test can know your sensor. A wrong V0, TC or range gives wrong temperatures that look plausible. The two limits to know:

  • 0 mV at the minimum (O-12). If V0 + TC x min is exactly 0 mV, a reading of 0 mV is a valid reading at the minimum, so a grounded or unpowered line is not caught by the window at any margin. Check at a known temperature at start-up, or add a pull-up on the board.
  • Positive outputs only (O-13). A part whose output goes below 0 V at its minimum, or whose slope is negative, is rejected by Validate().

Hardware

Not verified on hardware, and each item needs a bench:

Item Bench step
The model matches a real sensor Compare ReadRawPortMv() at two known temperatures with the table on Transfer function
The turn-on time and the unpowered output Troubleshooting, steps 1 and 2
The noise threshold of your board Step 3
The ESP32 ADC ceiling at 12 dB Step 4
The enable pin and the 200 Ω supply Power the sensor from the pin, read after the settle wait, and compare with a sensor on the supply rail