← Back to: eFuse / Hot-Swap / OR-ing Protection
Intro — Why Precision Sense
Precision sensing in the power path means wide common-mode range, high rejection (CMRR/PSRR + PWM immunity), true bidirectionality (incl. back-charge), low drift, and energy metrics wired to PG/FAULT/ALERT. Implementation flow: Rsense selection → front-end (INA/CSA or ΣΔ) → bandwidth & filtering → bidirectional thresholds → calibration & temperature mapping → layout (Kelvin/guard) & validation.
Vsense=I·Rs; P=I²Rs; target full-scale drop 10–50 mV.
Err≈Vos/(G·Vsense)+GainErr+TCRRs·ΔT; aim ±1–2%FS over temp.
P=V·I; E=Σ(P·Δt); align window with policing & telemetry cadence.
Front-end BW ≥ 5–10× signal; 80–100 dB ripple rejection via RC / sampling / ΣΔ SincN.
Topologies & Device Choices
High-side INA/CSA: wide VCM, high CMRR/PSRR, input clamps; avoid reverse-current saturation — best partner to eFuse/Hot-Swap. Low-side: lowest cost but ground noise sensitive; use only when return integrity is guaranteed. Isolated (ISO-amp / isolated ΣΔ): bandwidth/linearity vs latency/CMTI & isolated bias constraints. ΣΔ path: place PWM base into SincN notches; choose OSR for ENOB vs delay. Magnetic: zero burden, larger drift/non-linearity; great for protection adjunct or coarse metering.
BW ≥ 5–10× highest content; place a zero for PWM pole; ensure ADC sampling not aliased.
Do not confuse input range with VCM; check transient headroom & input clamps.
Validate negative I; many INAs saturate asymmetrically; add hysteresis to ALERT.
Kelvin to Rsense; guard sensitive nodes; keep PWM loops orthogonal to sense traces.
Rsense Selection (Value • Loss • Pulse • TCR)
Target a full-scale burden of 10–50 mV. Pick Rs=VFS/IFS, then verify P=I²Rs for continuous thermals and Epulse=∫I²Rsdt for surge windows. Control drift via low-TCR alloy shunts and symmetric copper for isothermal behavior. Kelvin pads are mandatory; keep dual-polarity headroom for bidirectional range.
VFS=10–50 mV → Rs=VFS/IFS; P=I²Rs.
Epulse=∫I²Rsdt; check against vendor derating curves.
ΔR/R≈TCR·ΔT; use low-TCR foil/shunt, keep copper isothermal.
Reserve margin for negative I; validate symmetry & hyst.
- Define IFS, Ipk(t), efficiency and min resolution → select VFS in 10–50 mV band.
- Compute Rs, check Pcont vs package rating and θJA.
- Integrate I²Rs over pulse window → verify Epulse with derating.
- Pick TCR & tolerance to meet total error; set symmetric copper for isothermal flow.
- Route true Kelvin sense traces; keep current path separate from sense path.
- Chasing resolution with high R → poor efficiency/overheat.
- Ignoring pulse rating & derating curves.
- Wrong Kelvin routing; mixing sense and current pads.
- Confusing tolerance with stability; TCU not characterized.
| Rs (mΩ) | VFS (mV) | Pcont (W) | Epulse (J/Δt) | TCR (ppm/°C) | Tol (%) | Pkg | θJA (K/W) | Kelvin | Notes |
|---|---|---|---|---|---|---|---|---|---|
| — | — | — | — | — | — | — | — | Yes/No | — |
Front-End Amplifier (CSA/INA)
Set G from VFS and ADC range, then ensure BW ≥ 5–10× signal content. Suppress PWM ripple via front-RC, phase-aligned sampling, or ΣΔ SincN notches. Verify VCM, input clamps, output swing, and reverse-current behavior. Budget error: ErrVos≈Vos/(G·VFS).
Vo=G·Vsense; check rail swing and G tolerance vs ADC headroom.
Vos, dVos/dT → Err≈Vos/(G·VFS) + GainErr.
CMRR/PSRR at PWM base & harmonics; clamp thresholds & recovery.
S/H kickback, ADC input RC; phase-aligned sampling window.
- Kelvin to shunt; short, symmetric diff routing; single-point return.
- Keep PWM switch loops orthogonal to sense; minimize input capacitance mismatch.
- Place input clamps close; verify recovery time under overload.
- Bode/step under various loads; measure phase margin.
- PWM injection (duty sweep) → Vripple in mVpp & dB.
- Negative current step; overload & recovery time; clamp behavior.
- ADC S/H kickback with target sampling phase; aliasing check.
Isolation Sensing
Choose between isolated amplifier and isolated ΣΔ by trading linearity, bandwidth, latency, CMTI, and isolated power noise. Align group delay with eFuse/hot-swap timing so metering and protection remain stable under high common-mode and long return loops.
Low latency, good linearity; sensitive to isolated supply ripple; drives ADC directly; check output swing & ADC load.
Bit-stream over isolation, digital SincN filtering; robust to analog noise, adds group delay; clean clocking required.
fBW ≈ (fclk/OSR)·k (k depends on Sinc order/definition).
Align group delay with protection loops (PG/FAULT); avoid sampling/hold conflicts.
Aim >50–100 kV/μs depending on switching edge rates; verify spec under load transients.
- ISO-amp → ADC: check swing, output drive, S/H kickback, anti-alias RC.
- ΣΔ Bitstream → MCU/PMBus: clock integrity, jitter, DRDY timing, metastability protection.
Pitfalls: isolated supply ripple injection; insufficient CMTI; mistaking analog BW for effective digital BW; delay not budgeted for loop stability.
- CMTI step tests; isolated PSRR sweep.
- Group delay vs PG/FAULT timing alignment.
- Temperature drift & post-surge zero shift checks.
ΣΔ Modulator & Digital Filtering
Use OSR and SincN to notch the PWM base/harmonics while keeping an effective bandwidth that supports power policing. Budget ENOB, RMS noise, and overload recovery so alerts remain stable and timely.
SNR≈6.02·N+1.76−10·log10( fs / (2·fBW) ); ENOB≈(SNR−1.76)/6.02.
fnotch=m·(fs/OSR); choose m to land on PWM base or key harmonics.
fBW≈(fclk/OSR)·k; higher order → deeper notches but longer delay.
- Pick Sinc order for notch depth vs group delay; window/EMA for stable power averaging.
- Time constants must match PG/FAULT semantics; avoid late alerts or chatter.
Pitfalls: notch misaligned to PWM; assuming nominal OSR equals true BW; slow overload recovery; clock coupling into power stage.
| OSR | Sinc Order (N) | fs | fBW | ENOB @ BW | RMS Noise | Group Delay | Notch Align | Overload Recovery | Notes |
|---|---|---|---|---|---|---|---|---|---|
| — | — | — | — | — | — | samples / μs | PWM base / Hn | ms | — |
Bidirectional Measurement & Alert Thresholds
Standardize semantics: PG=operable, ALERT=recoverable, FAULT=protective action. Define I⁺/I⁻ and Pavg/Ppeak thresholds with hysteresis and deglitch windows; integrate energy E=Σ(P·Δt) to support time/energy policing. Lock thresholds to eFuse Itrip/timers and ideal-diode ΔVtrip to avoid ping-pong.
P=V·I; E=Σ(P·Δt). Use Ppeak for instantaneous protection, Pavg for thermal/energy limits.
Threshold ±Δ with N-sample window (N·Ts). Choose N by ripple spectrum & response time.
ALERT_low → ALERT_high → FAULT. Latch/hold with timestamped event registers.
| I_pos_hi | I_pos_lo | I_neg_hi | I_neg_lo | P_peak_hi | P_peak_lo | P_avg_hi | P_avg_lo | debounce_samples | hold_time_ms | latch_mode | auto_clear |
|---|---|---|---|---|---|---|---|---|---|---|---|
| — | — | — | — | — | — | — | — | N | ms | Edge/Level | true/false |
| alert_source | efuse_action | ideal_diode_action | pg_state | notes |
|---|---|---|---|---|
| I_pos_hi | limit/fast-trip | — | ALERT→FAULT | — |
| I_neg_hi | reverse-limit | raise ΔVtrip | ALERT | avoid ping-pong |
Calibration & Temperature Compensation
Achieve 1–2%FS across temp with one/two-point production calibration, NVM-stored coefficients (with CRC), and online temperature compensation combining shunt TCR and AFE drift. Minimize thermal EMF by symmetric copper and true Kelvin routing.
ΔI/I ≈ TCRRs·ΔT + DriftAFE·ΔT + GainErr + OffsetErr/(G·VFS) + Quantization.
Asymmetric copper and gradients create μV offsets. Keep isothermal planes and shortest Kelvin paths.
{
"cal_version": "v1",
"timestamp": "UTC-ISO8601",
"offset_code": 0,
"gain_code": 1.0000,
"tcr_shunt_ppm": 25,
"drift_afe_uV_per_C": 0.5,
"temp_model": "linear|piecewise",
"temp_breakpoints_C": [-20, 25, 85],
"gain_table": [0.998, 1.000, 1.004],
"offset_table_uV": [-8, 0, 10],
"crc32": "0x--------"
}
- Two-point at production; verify residuals and across-temp drift.
- Write NVM with CRC & version; log timestamp and station ID.
- Online compensation with board/IC temperature; clamp extrapolation.
- Thermal EMF audit (IR camera, isothermal layout) and Kelvin routing.
PCB Layout & EMI (Kelvin • Guard • Return)
Guarantee electrical accuracy with geometry: enforce true Kelvin (+K/−K straight to AFE), short parallel differential, guard/“moat” ground to split power return, distance from switching nodes, symmetric copper for low thermal-EMF, and disciplined analog/digital partitioning.
- Kelvin leads: +K/−K run independently to the AFE pins; do not share power copper; avoid vias unless paired and symmetric.
- Short, parallel, same layer: differential pair with matched length and environment; minimize stubs; place anti-alias RC at the AFE end.
- Connector mapping: pinout aligns with reference plane and return stitching to prevent cross-zone currents.
- Moat/guard ground: separate power return loop from measurement/AFE island; use a single controlled bridge for reference continuity.
- Three-zone rule: power / analog / digital; crossings only through defined bridges or CM filters; no uncontrolled return across the sense pair.
- ESD/TVS returns: shortest loop to chassis/return without traversing sensitive nodes.
- Symmetric copper & isothermal islands: equalize temperature around Rsense; keep heat sources away from +K/−K routes.
- Via fields: balanced pairs near shunt pads for heat spreading without skewing Kelvin impedance.
- SW node clearance: maintain keep-out around switching edges; add “quiet” reference strip above/below the sense pair if needed.
- +K/−K do not share power copper
- Vias paired & symmetric
- Diff pair same layer/environment
- RC at AFE end
- Moat ground separates loops
- Single controlled bridge
- ESD loop shortest & independent
- No cross-zone return
- Isothermal copper around shunt
- Heat sources away from +K/−K
- SW keep-out respected
- Balanced via fields
Validation Matrix & Scripts
Execute end-to-end validation covering DC linearity, steps/saturation/recovery, PWM injection suppression, negative current/regen, energy integration error, across-temp/aging, and EMI/common-mode injection with explicit acceptance criteria and reproducible scripts.
- DC linearity/gain/offset at −40/25/85/105°C and 1/10/50/100% FS.
- Step/pulse tests: saturation entry & recovery, overshoot limits.
- PWM injection sweep: residual vs frequency to form suppression curve.
- Negative current/regen sign stability and false-trigger rate.
- Energy integration error vs reference under steady & dynamic load.
- EMI/CMTI robustness and event logging; aging drift delta.
uuid,temp_C,vin_V,load_A,waveform,duty,freq_Hz,sample_rate_sps,window_ms,osr,sinc_order,notch_align,gain,offset_uV,linearity_pctFS,enob_bits,residual_uV,recovery_ms,energy_meas_J,energy_ref_J,energy_err_pct,cmti_flag,emi_flag,alert_count,fault_count,pass_fail,notes
- Linearity ≤ X%FS; offset ≤ Y μV
- Recovery ≤ Trec; overshoot ≤ K%FS
- Suppression ≥ 80–100 dB
- |ΔE|/E_ref ≤ e%
- CMTI pass; no latch-up
- Synchronize clocks & sampling windows
- Log raw + derived metrics
- Produce suppression & recovery plots
- Emit per-test UUID and timestamp
Seven-Brand IC Shortlist (Shunt/Isolated/ΣΔ/Power Metering)
Shortlist focuses on production-proven parts for precision current/power sensing. Each pick states why it’s favored (CMRR, drift, PWM rejection, bandwidth, isolation, or metering features) and key caveats (VCM, latency, layout/Kelvin, thermal).
Why: Exceptional PWM common-mode step rejection in hot-swap and motor PWM rails; low Vos/drift; wide G options.
Notes: Respect input RC at AFE and keep Kelvin tight; verify VCM headroom.
Why: High bandwidth for transient/peak capture with good CMRR; suits fast eFuse trip correlation.
Notes: Don’t over-filter; confirm stability with source impedance.
Why: Integrated shunt voltage + bus voltage + power calc; quick bring-up; logging-friendly.
Notes: Sample rate/averaging trade-off vs ripple; calibrate for Rsense value.
Why: Multi-rail metering with alert thresholds and summation; good for system policing.
Notes: Channel coupling vs sample cadence; align ALERT deglitch with policy.
Why: Reinforced isolation, high CMTI for noisy hot-side measurement; linear and low drift.
Notes: Isolated supply noise control; layout return bridge discipline.
Why: Metering-grade ENOB with SPI, flexible OSR/Sinc for PWM notch alignment.
Notes: Latency affects protection timing; retune OSR when migrating.
Why: Industrial temp, decent CMRR, cost-effective for multi-rail boards.
Notes: Verify gain options vs VFS target (10–50 mV).
Why: Handles charge/discharge paths; compact for space-constrained rails.
Notes: Layout Kelvin and short RC to tame ripple.
Why: Energy metering front-end, ΣΔ bitstream to MCU; good linearity for power calc.
Notes: OSR/filtering defines latency; isolate digital return properly.
Why: Integrated metrology with power/energy accumulation and interfaces.
Notes: Evaluate front-end impedance vs Rsense; calibration flow is required.
Why: High-accuracy cell/pack sensing, robust diagnostics; good for bidirectional pack paths.
Notes: Digital filtering latency; align thresholds with protection timing.
Why: System power monitors integrated with safety PMICs; simplifies logging and fault paths.
Notes: Confirm monitor bandwidth vs DC/DC switching spectrum.
Why: Mature current/voltage/power monitoring with alerts; solid across temp.
Notes: Average settings vs response; calibrate Rsense and bus voltage.
Why: Better accuracy and configurability; good for policy policing.
Notes: Verify ADC rate vs ripple; ALERT semantics mapping.
Why: Metering-class resolution for precision shunt measurement; SPI readout.
Notes: Layout analog ground island; filter latency management.
Why: Simple, low-offset, multiple gains; good availability for volume.
Notes: Check VCM vs rail; bandwidth vs ripple requirement.
Why: Build discrete INA stages or filters with low drift; automotive options.
Notes: Input common-mode and output swing headroom.
Why: Low offset/drift with good gain options and temperature range; compact packages.
Notes: Keep RC at input close; validate gain-bandwidth vs ripple.
Why: Multi-channel ΣΔ front-end with energy-friendly specs; SPI control.
Notes: Clocking & OSR plan; isolation and ground bridges.
Why: Built-in power/energy accumulation and alerts; fast to deploy.
Notes: Calibrate gain/offset vs Rsense; check sample averaging.
Why: High current, low dissipation, galvanic isolation advantages; good linearity after calibration.
Notes: Thermal drift model; mechanical alignment critical.
Why: Wider bandwidth for transient capture; integrated conditioning.
Notes: Shielding and field shaping near busbars.
Why: Reinforced isolation variants, digital configurability fits metering/policy loops.
Notes: Cal/linearization; evaluate delay vs protection needs.
Cross-Brand Alternatives & Migration
Priority order: architecture/common-mode scope → accuracy & drift → bandwidth & PWM suppression → interface/pin-map. For ΣΔ↔analog front-ends, recompute OSR/Sinc and re-align notch vs PWM base; adjust ALERT deglitch to match new latency. For Rsense, preserve R·Imax (10–50 mV VFS) and pulse power.
| Source Part | Target Brand/Part | Why Comparable | Retest Focus |
|---|---|---|---|
| TI INA240 (PWM-robust CSA) | ST TSC213 / Microchip MCP6C04 | High-side CSA with low Vos/drift, similar gain options | PWM suppression depth; recovery after saturation |
| TI AMC1301 (isolated amp) | Renesas ISL26134 front-end + isolator path | Isolation + precision FE yields similar signal chain | CMTI; isolated supply noise; latency budget |
| TI ADS131M04 (ΣΔ ADC) | ST STPMS2 / Microchip MCP3913 | ΣΔ metering front-ends with SPI/bitstream output | OSR & Sinc notch realignment; energy calc scaling |
| TI INA226 (power monitor) | Renesas ISL28023 / onsemi NCS199A1 + ADC | Similar current/bus/power registers or simple chain | ALERT semantics; averaging settings vs ripple |
| Melexis MLX91216 (magnetic) | TI shunt path (INA240) + low-R shunt | When zero-burden is not mandatory; improve precision | Loss/thermal; calibration redo; EMF and layout changes |
- Architecture/Common-Mode: VCM min/max, bidirectional range, isolation & CMTI; ΔVtrip and ideal-diode interactions.
- Accuracy/Drift: Vos, Vos/°C, GainErr, INL/ENOB (ΣΔ), calibration method and NVM format.
- Dynamics: −3 dB BW, step recovery, PWM suppression depth at target frequency; ΣΔ latency budget.
- Semantics: ALERT/FAULT/PG levels, deglitch & latch behavior; energy registers scaling.
- Mechanicals/Thermal: Pin-map, package θJA, Rsense footprint & TCR, Kelvin pad compatibility.
BOM Remark — Precision Sense Chain 1) Architecture & VCM: Replacement must support VCM ≥ __ V, bidirectional current range ≥ __ A, CMTI ≥ __ kV/μs (if isolated). 2) Accuracy: Vos ≤ __ μV, Vos/°C ≤ __ μV/°C, GainErr ≤ __ %, ENOB ≥ __ bits (ΣΔ). 3) Dynamics: BW ≥ __ kHz, PWM suppression ≥ __ dB @ __ kHz; ΣΔ OSR/Sinc = __ / __; latency ≤ __ μs. 4) Semantics: ALERT deglitch = __ samples, latch = __ ms; energy registers scaling = __. 5) Shunt: R = __ mΩ, I_max = __ A (V_FS ~ 10–50 mV), pulse power ≥ __ W; TCR ≤ __ ppm/°C; footprint & Kelvin pads compatible.
FAQ — Precision Current/Power Sense
Keep Rsense burden near 10–50 mV, align bandwidth and filtering to your PWM spectrum, and map ALERT semantics to PG/FAULT without ambiguity.
How do I pick Rsense so burden voltage is low yet accuracy is high?
Choose R to place full-scale burden around 10–50 mV. Below 10 mV, offset and drift dominate; above 50 mV, I²R loss and heating raise error. Validate pulse power and TCR. Use Kelvin pads, symmetric copper, and short differential sense leads. Calibrate gain with the actual shunt value and temperature range.
High-side vs low-side: when do I need isolation instead?
Use high-side for accurate load current independent of ground shifts; low-side is cheaper but injects ground disturbance. Choose isolation when VCM exceeds amplifier limits, when safety domains differ, or when CMTI is critical near fast switching nodes. Isolation also breaks ground loops and protects the controller from fault energy.
How do I keep PWM ripple from corrupting readings?
Place a small RC at the amplifier input and keep it close to the pins. Ensure amplifier bandwidth is high enough to avoid slewing while the RC attenuates switching edges. For ΣΔ paths, align Sinc filter notches to the PWM base or its harmonics. Verify residual ripple versus frequency with an injection sweep.
What CMRR and input range do I need in a hot-swap path?
Target very high CMRR across the switching spectrum, not just at DC. The input must tolerate fast common-mode steps during plug-in, gate ramp, and fault interruptions. Ensure headroom for the maximum bus voltage, transient overshoot, and reverse conditions. Validate saturation and recovery times with representative inrush and trip events.
When is a ΣΔ modulator better than a classic CSA + ADC?
Use ΣΔ when you need high effective resolution, stable digital filtering, and inherent antialiasing. It excels for energy metering, slow to medium bandwidth control, and noisy environments. Account for latency versus protection timing. Align OSR and Sinc order to place notches on the PWM base frequency to suppress ripple efficiently.
How do I implement true bidirectional thresholds with hysteresis?
Define separate positive and negative current thresholds with symmetric or application-specific margins. Add hysteresis bands to avoid chatter near zero. Deglitch over N samples to reject spikes. Map ALERT to early warnings and FAULT to latched events. Recalculate average windows so energy and direction decisions remain consistent across PWM duty changes.
How should I calibrate across temperature without long test time?
Use a two-point or one-point plus modeled TCR calibration. Store shunt value, gain, and offset in NVM with a temperature tag. During production, sweep a minimal set of currents at two temperatures, then interpolate. At runtime, apply temperature compensation from a local sensor and periodically re-validate at known operating points.
What layout mistakes add thermoelectric drift or EMF offsets?
Unequal copper around the shunt, long asymmetric sense traces, and vias only on one lead create thermal gradients and Seebeck voltages. Avoid routing over heat sources. Keep sense leads short, parallel, and same layer. Pair vias symmetrically. Use isothermal copper islands and Kelvin pads to minimize junction temperature differences.
How do I log energy robustly with limited MCU bandwidth?
Integrate power over fixed windows using accumulated sums rather than high-rate raw streams. Use moving averages matched to filter latency. Compress to 32-bit scaled integers with explicit units. Log window start time, duration, and saturation flags. For transient capture, add event-driven bursts while preserving a low-rate baseline for long records.
Can I share one ALERT/FAULT pin between metering and eFuse?
Yes, if levels, polarity, and timing are explicitly defined. Use wired-OR with pull-up and ensure deglitch windows avoid masking faults. Differentiate events via status registers or priority encoding. Keep FAULT latched until handled; keep ALERT re-armable. Document semantics so firmware never confuses early warnings with hard protection trips.
What changes when I move from shunt to magnetic sensors?
You gain near-zero burden and galvanic separation but must model linearity, offset drift, and mechanical alignment. Magnetic pickup is sensitive to geometry and nearby fields. Temperature compensation and calibration become more important. Latency and bandwidth can differ; re-verify threshold timing, noise floors, and scaling before enabling protection policies.
How do I validate small-signal accuracy and large-pulse survivability?
Split tests: first, DC accuracy and linearity across temperature at low currents; second, step and pulse injections up to worst-case. Measure saturation entry, recovery time, and residual error. Validate energy integration versus a reference meter. Add EMI and common-mode bursts. Record conditions, results, and pass criteria in a repeatable matrix.