← Back to: Battery Charging / Gauging / Protection / BMS
Role in the Charging Branch
This chapter explains why multi-point temperature sensing, per-channel linearization, and fault-aware readout must live inside the charging branch rather than in a generic, slower BMS layer. The charging loop needs fresh, clean, and synchronized temperatures to drive charge current limiting, JEITA-like derating, or even a short pause.
Key idea: the charger / power-path / USB-C controller reacts in tens to hundreds of milliseconds. If multiplexed NTC/RTD data arrives slower than that or arrives unlinearized, the charge current will oscillate and look unstable.
In 2S–6S packs we often have more hot spots than the charger can monitor directly: cell cluster, FET zone, balancing resistors, connectors or harness. Multiplexing helps us bring 8–12 sensors into 1–2 monitor inputs, but it also introduces channel-switch latency, shape differences between NTC and RTD, and false over-temperature reports. This page is exactly for handling those effects before the charging logic consumes the data.
What this chapter solves
- Charger IC exposes only 1–2 temp pins, but the pack has 8–12 NTC/RTD probes.
- Thermal rise is non-uniform: cell area warms slowly, FET area and balancer can heat up very fast.
- Existing JEITA flow expects just one temperature, but now we must feed it multiple zones without triggering false derating.
- One sensor or one MUX channel may fail (open/short). That must not shut down charging for the whole pack.
This chapter intentionally does not cover: NTC package selection, thermal simulation, vehicle-level temperature standards, or pack-FET action logic.
Sensing Front-End & Mux Topologies
Multi-point temperature in charging BMS can be reduced to three engineering steps: how to excite the sensors, how to switch between many probes, and how to read them fast enough so that charge current does not shake. This chapter compares the three most practical topologies for engineers working on 2S–6S packs.
If the MUX switch time + settling time is longer than your charging derating interval, you will see false over-temperature and unstable current. That is why topology selection and grouping matter.
The charging branch often has to mix common 10 kΩ NTCs with a few RTD probes placed far from the board. Mixing is allowed, but excitation and linearization cannot be shared blindly. Each channel must carry its own metadata so later chapters (open/short detection, thermal map) can trust the value.
Topology A: Analog MUX + ADC
Cheapest and very flexible. Needs per-channel settling time (RC, long traces). Best when you can group sensors by speed.
Topology B: I2C temp monitor (multi-channel)
Integrated ADC and registers. Faster to integrate with charger MCU. Beware of brand-specific register maps and missing open/short detect.
Topology C: Several small 2–4ch devices
Good for small-batch builds and unstable lead times. Requires address planning and multi-device polling.
This chapter intentionally does not cover isolation ADCs, I2C bus arbitration, full PI analysis, or material-level NTC-vs-RTD selection.
Linearization & Per-Channel Calibration
The charging branch consumes temperature, not raw resistance or ADC codes. Once we multiplex 8–12 NTC/RTD probes into 1–2 inputs, every channel may have a different β value, cable length, or thermal environment. If we push these mixed readings straight into a single JEITA curve, we will derate at the wrong time. This section builds a per-channel linearization path so that all sensors land on the same temperature coordinate before charging decisions are made.
Goal: stable charge decisions (limit / derate / pause) across all temperature points, not lab-grade 0.1 °C accuracy. Each channel gets its own method, coefficients, and offset.
What this chapter solves
- Mixed 10 kΩ NTC and RTD under the same MUX: can we use one code path? (Yes, if channel metadata is explicit.)
- Remote NTC reads slightly higher because of cable/placement: should we offset it? (Yes, at the charging layer.)
- Procurement swaps the sensor brand: how not to break all temperature tables? (Use a channel-level config table.)
- How to express channel → zone → derating as a simple config so small batches can maintain it.
Structure
3.1 Temperature goal for the charging branch
3.2 NTC linearization methods (LUT / segmented / Steinhart–Hart)
3.3 RTD simplification (PT1000 in automotive)
3.4 Channel-level calibration table
3.5 Syncing calibration with the charge state machine
3.6 Feeding results to open/short detection (always linearize first)
Channel calibration table (concept)
channel_id → sensor_type → lin_method → coeff_a/b/c → offset → zone_map. Change sensors? Just update this table, leave derating logic intact.
channel_id: 0..11sensor_type: NTC_10K / NTC_47K / RTD_1K / RTD_1K-Ptlin_method: LUT / SEG / SHoffset: +1.2 °C for long cableszone_map: CELL / FET / CONN
This section does not cover factory-level NTC grading, MCU flash layout for large tables, or cloud/remote calibration. Those belong to manufacturing or higher BMS layers.
Open/Short Detection & Sensor Health
The biggest risk in a multiplexed temperature front-end is not precision, but fake temperature. A single open probe, a shorted line, or a still-settling MUX channel can look like over-temperature and wrongly pull down charging current. This chapter adds a fault-aware layer on top of the linearized temperatures, so that one bad channel does not stop the whole charge.
Strategy: detect the bad channel → lock it → continue charging with reduced margin → report upstream. Do not take the entire BMS page down.
What this chapter solves
- MUX just switched, ADC not settled, looks like an open sensor — how to avoid false faults?
- Probe wire broken — is it reported as very high or very low? (must be unified)
- Only 1 out of 10 channels is faulty — can charging continue? (yes, in degraded mode)
- Balancing heats up a nearby zone — should we re-sample before locking the channel? (yes)
This section does not define CAN/UART reporting, host-side logging, or vehicle-level DTCs. It only ensures the charging branch stays alive when one sensor fails.
Thermal Map & Charge-Derating Hooks
The charging branch does not need 10–12 individual sensor names. It needs 3–4 thermal zones that map directly to charge current decisions: Cell zone (safety first), FET / power zone (fastest to overheat), Connector / harness (current dependent), and optionally Ambient. This section shows how to group multiplexed NTC/RTD readings into those zones and how to trigger derating without rewriting the existing JEITA flow.
Goal: “many sensors → few zones → existing charging logic”. Derate only on the zone that is hot; keep charging alive when other zones are normal.
Why we need a thermal map
Multiplexing adds reach, but it also adds noise and different heating rates. If every sensor can directly trigger derating, the charge current will look jumpy. By mapping sensors into zones, we only act on area-level heat, which is what charging actually cares about.
Zone modeling methods
Board/placement-based
Group by PCB area: cell-sense corner → Cell zone; high-side FET / inductor area → FET zone; input / connector → Connector zone.
Function-based
Anything tied to balance, charge switches, power-path → FET zone; sensors glued to cell cans → Cell zone.
Heat-rate-based
Use ΔT/Δt to detect fast-rising points; temporarily increase their polling rate without touching other zones.
This chapter does not cover full-system thermal CFD, fan/cold-plate control, or external thermal-ECU protocols. We stay in the charging branch only.
IC Selection & Cross-Brand Shortlist (7 vendors)
This section is for purchasing and for engineers who need real part numbers. It lists ICs that can actually sit in the charging branch to read multi-point NTC/RTD, provide open/short detection, and report clean temperatures to the charger. We group them in three layers: devices with multi-temp inputs already, standalone multi-channel temperature monitors, and AFEs whose temperature pins can be reused.
Rule: same channel count → same interface (I²C / SMBus) → same linearization/fault ability → then pick package / AEC-Q hundred series. BOM must explicitly say “prefer parts with open/short detect”.
Selection dimensions
- Channel count (4, 6, 8, 9) and whether it can be paralleled.
- Linearization ability (NTC LUT, remote diode, series-R correction).
- Fault detection (open/short, out-of-range, sensor missing).
- Interface (I²C / SMBus) for minimal firmware change.
- Automotive readiness or close family with AEC-Q100 members.
Vendors & example parts
TI
- TMP468 – up to 8 remote + 1 local, SMBus
- TMP464 – 4-channel remote temp monitor
- TMP423 – remote diode monitor for hot spots
- TMP451-Q1 – automotive, remote sensor
- BQ25792 – charger with temp input, needs external monitor for 8–12 pts
ST
- L9963E – automotive battery monitor, up to 7×NTC
- L9961 – BMS front-end, temp inputs
- STTS751 – I²C temp sensor, can cover slow/ambient points
NXP
- MC33771B – BMS AFE with multiple temp inputs
- MC33772C – cell monitor + temp
- MPC5775B – BMS MCU side, can poll multiple monitors
Renesas
- ISL78714 – 14 cells + temp inputs, good for “reuse temp pins”
- ISL94216 – battery pack monitor with temp sensing
- RAA489204 – charger controller, needs external multi-temp
onsemi
- NCT214 – remote temperature monitor (I²C)
- NCT72 – temp sensor, use multiple with address offset
- NCT1008 compatible – for small builds, parallel 2–3 pcs
Microchip
- EMC1414 – 4-channel temp monitor, SMBus
- EMC1815 – multi-channel with fault
- MCP9808 / MCP9844 – for ambient/connector points
Melexis
- MLX90614 – remote IR temp, use as “difficult-to-wire” point
- MLX90632 – compact IR sensor, connector/harness spot
- MLX90327 – automotive sensing family for special locations
This chapter intentionally excludes low-value or non-automotive sensors, non-7-vendor parts, and market pricing. It stays focused on parts that can sit next to the charger and feed the thermal map.
Small-Batch Procurement & BOM Notes
This section explains how to trade off channel count, linearization, and open/short detection when parts are not fully available. It is written for small-batch / pilot builds, but it still keeps the whole page inside the charging branch of the BMS. Do not downgrade this block to a “single NTC input charger IC”, because the previous chapters (linearization, fault, thermal map) all expect multi-point temperature data.
Rule for small batches: you may reduce one dimension (e.g. channels) but you must keep the other two (linearization + open/short) so the charging derating does not start to chatter.
Typical small-batch scenarios
A: Channel not enough
You only got a 4-ch temp monitor but the design needs 8–12 pts. Solution: place two identical devices on the same I²C/SMBus, give them different addresses, and map Cell zone to device #1, FET/connector to device #2.
B: Brand changed
Original design used TI TMP468 but only Microchip EMC14xx or onsemi NCT family is in stock. Keep the interface and alarm concept the same, push fine linearization back to the charger firmware.
C: No open/short detect
You have a temp IC that reports temperature but no remote-fault bits. Keep the hardware, but add MCU-level re-sampling + time window so that “one bad channel” does not stop charging.
BOM remark templates
These sentences are meant to be copy/pasted into ERP / WordPress product pages / download PDFs. Keep them in English.
- Template 1: Temp MUX must support per-channel open/short detect, do not replace with single NTC input charger IC.
- Template 2: Channel linearization is handled in charger firmware; keep the sensor type and channel assignment unchanged across alternatives.
- Template 3: If only 4-channel monitor ICs are in stock, populate two devices on the same bus with different addresses to reach 8+ channels.
- Template 4 (lead-time): Prefer AEC-Q qualified parts; industrial grade is allowed for pilot lots if interface and fault behavior match.
This chapter intentionally excludes pricing, distributor lists, and grey-channel sourcing; it focuses on technical equivalence in the charging branch.
Request a Quote
Frequently Asked Questions
All questions below are scoped to this page: multi-point NTC/RTD MUX and readout for charging-focused BMS designs. We do not drift to generic “what is BMS” topics. Two questions (11, 12) are tuned for small-batch and cross-brand situations to help procurement convert.
1. How many NTC channels can I safely multiplex before charge-derating becomes too slow?
As a rule, your total round-trip polling time for all channels should be shorter than the charging loop’s sampling/decision time. 1 s / 8 channels is usually safe; beyond 10–12 channels on a slow bus, split them into two groups or increase the polling rate.
2. Do I need different linearization tables when mixing NTC and RTD on the same MUX?
Yes. Keep per-channel metadata: sensor type, coefficients, and offset. The MUX only switches the source; the charger must normalize each channel into the same temperature scale before derating.
3. How do I distinguish a real over-temperature from an open NTC that reads full scale?
First linearize, then compare. Add 2–3 re-samples in a short window, and flag as “fault” only if the channel stays at max with very low noise. A real hot zone typically rises gradually or in sync with charging/balancing.
4. Should the charger pause or just reduce current when one thermal zone trips but others are normal?
Derate first, pause later. If only FET/power zone is hot, reduce current by a fixed step. Pause or alert only when cell zone or multiple zones are above limit.
5. What sampling interval is recommended when cell balancing is active and causes local heating?
Normal operation: about 1 s per full sweep. During balancing or fast heat events: 300–500 ms, and temporarily prioritize the affected zone’s channels.
6. Can I share the same MUX for temperature and other analog signals without corrupting the readings?
You can, but add settling time after each MUX change, and poll temperature before low-level analog channels. Keep the charging-temperature path highest priority.
7. How do I report per-channel sensor faults to the BMS host without blocking charging?
Report two things: the final zone temperatures (for charging decisions) and a fault bitmap listing which channels were excluded. Charging should derate, not stop, when only one channel fails.
8. What’s the best way to map 10–12 physical sensors into 3–4 thermal zones for JEITA-like derating?
Map by function: Cell zone for anything near cells, FET zone for switching and power-path, Connector zone for high-current terminals, Ambient as reference. Do not assign one sensor to multiple zones.
9. Can I connect remote NTCs over long wires without picking up charger switching noise?
Yes, use twisted pair, series resistor + RC at the ADC/MUX side, and route away from high dV/dt nodes. If noise is still present, increase the per-channel settling time.
10. How do I calibrate each channel if the NTCs have different β values or tolerances?
Keep a channel-level table in firmware: channel_id, sensor_type, coeff_a, coeff_b, coeff_c, offset.
Do not overwrite the charger’s global temperature table just because one sensor was replaced.
11. Is it okay to replace a multi-channel temperature monitor IC with several single-channel parts in small batches?
It is acceptable for pilots if you keep the same bus, offset the addresses, and add software open/short detection. Mention this explicitly in the BOM so purchasing does not downgrade later runs.
12. Which parameters must stay identical when switching brands (TI → ST → NXP) to avoid charge-derating mis-triggers?
Keep the sampling rate, temperature resolution/step, alarm polarity, and fault-latch behavior the same. If these change, the charging current will fluctuate even though the hardware is fine.
