Skin Temperature Sensor Design for Rings and Bands

Concept render

Datasheet accuracy describes the chip at its board location. The TMP117 holds ±0.1°C maximum only from -20°C to 50°C, and it measures the board where it is mounted. In a ring or band, the thermal path to skin and to ambient air adds further error, so describe the output as a trend for general wellness.

A datasheet says ±0.1°C, and the product still reads differently depending on fit, shell and room temperature. This guide separates the chip from the path around it. It uses the TMP117 datasheet as the only source for numbers, then gives formulas you can rerun with your own values. It contains no Duobay measurements and no price. For how skin temperature fits with the other sensors in a ring or band, see wearable sensors: types, power and placement.

This is an engineering write-up for general wellness products. It is not a medical device guide, and it does not cover diagnosis, treatment or clinical thermometry.

What does the TMP117 datasheet actually promise?

The table lists the numbers this article uses. Every row comes from the TI TMP117 datasheet, as of October 31, 2026.

The datasheet states that the part measures the temperature of the board location where it is mounted. The sensor accuracy therefore says nothing about the skin-to-sensor path. That path is your design.

How does TI tell you to lay out a surface sensor?

The datasheet gives concrete layout rules for surface-temperature sensing, and it references a wearable layout app note (SNIA021). In summary: use a thin PCB, a copper plane, thermal paste to the object, minimal copper wiring on the top layer, and thermal isolation from the surrounding air. It warns about "leakage" to ambient air and says to cover the sensor with thermally isolating foam or tape (source, as of October 31, 2026). For a ring, the shell, the potting and the battery sit around the sensor, so each of those is part of the thermal path. Whether your stack-up meets these rules is something you check on your own board.

How far off can a good sensor read inside a ring shell?

Treat the sensor as a node between two thermal resistances. One connects skin to the sensor. The other connects the sensor to ambient air. At steady state the reading sits between skin and ambient, and the offset from skin is:

offset = (T_skin − T_ambient) × R_ss / (R_ss + R_sa)

T_skin, T_ambient   °C
R_ss                skin-to-sensor thermal resistance, °C/W
R_sa                sensor-to-ambient thermal resistance, °C/W
offset              °C, reading below skin temperature when T_skin > T_ambient

The table uses a skin-to-ambient gap of 10°C and three pairs of resistances. The gap and every resistance value below are illustrative assumptions for you to replace. They are not measurements of any product. Only the formula is general.

Check: case B is 10 × 10 ÷ (10 + 100) = 0.91°C. Even with a ±0.1°C sensor, an assumed path like case B reads about 0.91°C low, and the offset grows with the gap between skin and room. A cold room moves the reading even when the wearer's skin has not changed. The lesson follows from the formula: keep R_ss low and R_sa high, which is what the TI layout rules aim at.

How fast does the reading follow a change?

A first-order estimate is tau = R_th × C_th, with R_th in °C/W and C_th in J/°C, which gives seconds. The datasheet thermal mass is in mJ/°C, so convert before multiplying.

Check: 133.2 × 0.0008 = 0.107 s, and 70.7 × 0.0051 = 0.361 s. A first-order system reaches about 95% of a step after three time constants (1 − e⁻³). The sensor alone is fast. The assembly around it sets the lag you ship, and the assembly row is an assumption. Measure the real settling time on your build, as described in the test table below.

How does sampling interval relate to self-heating?

The datasheet says conversion dissipates power that heats the device, that the effect depends on power and board size, and that a 1-second conversion cycle lets the device cool (source, as of October 31, 2026). It lists a 15.5 ms one-shot conversion time and 3.5 µA at 1 Hz. Duty cycle is conversion time divided by cycle period:

duty = t_conv / t_cycle

t_conv    15.5 ms (one-shot, datasheet)
t_cycle   your sampling period

Check: 15.5 ÷ 1000 = 0.0155, and 15.5 ÷ 60000 = 0.00026. Duty is a design lever, but average current also includes the standby current, which this article did not capture. Read it from the datasheet for your mode. How much the reading shifts from self-heating on your ring board is not published, so it has to be measured on your board.

How do the error terms add up?

An error budget lists each source and combines them. Root-sum-square (RSS) is a common combination when the terms are independent: total = sqrt(e1² + e2² + e3² + e4²). Only the sensor term below is from the datasheet. The other terms are illustrative assumptions for you to replace.

Check for case A: 0.1² + 0.2² + 0.24² + 0.05² is about 0.112, and the square root is 0.33°C. Two cautions follow from the math. First, the chip term is small next to the ambient term in these cases, so a better chip changes the total very little. Second, the ambient term is a systematic offset that moves with room temperature, so RSS can understate it. Treat the total as a way to see which term dominates, and replace the assumptions with bench data.

What can you claim about the output?

The TMP117 datasheet says the chip is designed to meet ASTM E1112 and ISO 80601 requirements for electronic patient thermometers (source, as of October 31, 2026). That is a statement about the chip in a clinical thermometer design. This article does not describe the scope of ISO 80601. A ring or band is a different product with a different thermal path, and it does not inherit the sentence.

For a general wellness product, write about change relative to the wearer's own baseline over time, and keep copy free of diagnosis, treatment and medical language. This is an engineering summary. It is not legal or regulatory advice. Have your regulatory advisor review the final wording.

What about NTC thermistors versus digital sensors?

This article gives no figures for other digital sensors or NTC thermistors. Read the current datasheet of the exact part. What you can do now is compare on your own board, in the same shell, with the same fit. The test table below has a row for it. When you read each datasheet, look for which error terms it lists and under what conditions, and for what the read circuit adds.

What are the common pitfalls?

These follow from the datasheet facts above. They are engineering reasoning about how the numbers behave. None describes an incident at any company.

• Reading the chip accuracy as the product accuracy. The datasheet says the TMP117 measures the temperature of the board location where it is mounted. The ±0.1°C figure therefore covers the chip at that spot. The path from skin to that spot is outside it.
• Quoting the headline band outside its range. The ±0.1°C maximum applies from -20°C to 50°C. The datasheet gives ±0.15°C over -40°C to 70°C and ±0.2°C over -40°C to 100°C. If your product spec reaches those temperatures, use the wider band.
• Leaving a path to the air. The datasheet warns about "leakage" to surrounding air and recommends thermal isolation. In the offset formula, a low sensor-to-ambient resistance pulls the reading toward air temperature. Engineering inference: a ring shell that conducts to the outside lowers the reading in a cold room.
• Heat that arrives on copper. The datasheet recommends minimal copper wiring on the top layer for surface sensing. Engineering inference: traces that run to warmer parts of the board can carry heat into the sensor and shift the reading with activity.
• Self-heating on a small board. The datasheet says conversion dissipates power that heats the device, and that the effect depends on power and board size. Engineering inference: a ring board is small, so verify the effect on your own assembly instead of assuming the datasheet condition applies.
• Assuming package choice only affects size. The two packages have different thermal resistance and thermal mass (133.2 °C/W and 0.8 mJ/°C against 70.7 °C/W and 5.1 mJ/°C). Those values give different sensor-only time constants, so the package choice affects how fast a reading follows a change.
• Letting a chip sentence turn into a product claim. The datasheet says the chip is designed to meet ASTM E1112 and ISO 80601 requirements for electronic patient thermometers. That sentence describes the chip. A ring or band built on it does not inherit it, and the product copy must not echo it.

What should you test on your own board?

The datasheet gives no fit, shell or ambient behavior for your product, and we have not run these tests for you. Use a calibrated reference thermometer with a calibration record. Pass lines are set by you from your product requirements, and you record the basis for them.

Write each pass line down before you run the test. A threshold chosen after you see the result tends to land just under the result.

What should you think through before choosing?

Take this list to your team and to any supplier who proposes a sensor or a shell.

• Which exact claim will the product make about temperature, and does it describe a change over time or an absolute reading?
• Where does the sensor sit relative to the skin contact point, and what material is between them?
• What is between the sensor and the outside air, and does anything conduct heat to it from the shell?
• Which package are we using, and have we computed its sensor-only time constant from the datasheet values?
• What sampling interval does the battery plan assume, and what duty cycle does that give for the conversion time?
• Which assumptions in our error budget are measured on our own board, and which are still guesses?
• What temperature range does the product specification cover, and which datasheet accuracy band applies to it?
• Is there a calibrated reference thermometer with a calibration record for the bench tests?
• Have we read the datasheet of any NTC or other candidate part, and which error terms does it list?
• Does any text in our app, packaging or website sound like a diagnosis, a treatment or a medical measurement?

If you are planning a ring or band with temperature sensing, tell us about your device. For how this fits our health wearable work, see health wearable development.

Sources

The page below was accessed on 2026-10-31. Datasheets are revised, so confirm figures on the live document.

• Texas Instruments, TMP117 datasheet: https://www.ti.com/lit/ds/symlink/tmp117.pdf (accessed 2026-10-31)