Compared with a tablet

The detachable adds a dock accessory that must be safely powered, detected, and debounced. That introduces contact resistance drift, attach/detach bounce, and ESD at exposed pins.

Compared with a laptop

The detachable has weaker ground continuity and a more fragile return path at the interface. Touch/pen sensing is also more sensitive to reference shift driven by charger noise or hot-plug transients.

Contract A — Pogo/Magnetic Dock

Carries: dock power, low-speed I/O (I²C/GPIO/scan), ID/detect.

Breaks first: contact bounce, resistance drift, ESD at exposed pins.

Protect roles: current limit/load switch, ESD arrays near entry, robust return path.

Measure first: Vdock droop at pins + attach debounce statistics + I/O glitch capture.

Contract B — Touch & Pen Sensing

Carries: high-impedance sensing, scan drive/receive, pen coupling path.

Breaks first: reference shift and narrowband interference during charge/dock events.

Protect roles: layout isolation, quiet rails, controlled return paths (not “more filtering”).

Measure first: raw noise/jitter vs state (tablet/dock/charge) + spectrum of the disturbance.

Contract C — USB-C Power Path

Carries: VBUS power, role-level PD control, power switching between sources/sinks.

Breaks first: inrush droop, reverse current fights, ground/ESD injection through shield.

Protect roles: power-mux + reverse blocking + eFuse/load switch + controlled inrush.

Measure first: VBUS plug-in transient + system rail stability + reset reasons/log timestamps.

Contract D — ULP Sensors & Wake

Carries: always-on sensing and wake interrupts (Hall/IMU/ALS).

Breaks first: leakage and false wakes that destroy standby targets.

Protect roles: strict rail gating, interrupt hygiene, power-on default states.

Measure first: rail-by-rail current split + wake cause counters + time-correlated event logs.

State A — Tablet-only

Dock is disconnected. Dock domain should be off or high-impedance. Touch/pen performance is the baseline reference for later comparisons.

• Evidence: baseline touch raw noise + pen jitter baseline (if supported). Used to isolate charger/dock coupling.
• Budget: standby current dominated by AO domain; scanning is periodic and bounded.

State B — Dock-attached

Dock is attached and powered through a controlled enable sequence. Debounce is a system feature (electrical + event gating), not a delay.

• Evidence: Vdock minimum during attach + attach event counts. Detect bounce shows up as repeated attaches.
• Budget: additional dock rail current + I/O event activity; attach spikes must stay inside the droop window.

State C — Charger-attached

VBUS enters and the power path arbitrates sources. The most common touch/pen failures are reference shift and narrowband coupling during switching.

• Evidence: VBUS plug-in transient + system rail droop + raw noise delta (charger A/B). Always compare to Tablet-only baseline.
• Budget: charging current peaks + switching activity; the sensing reference must remain stable.

State D — Sleep / Modern Standby

Only wake-critical functions remain active. Standby failures are typically caused by default leakage and false wakes, not “battery problems”.

• Evidence: rail-by-rail current split + wake source counters. False wake frequency should be measurable.
• Budget: AO domain defines the floor; every wake source must be deterministic and quiet.

Pin partitioning (Power / Signal / Detect / Shield)

Group pins by function and stability. Power/ground should establish a reference before signals are allowed to toggle. Detect/ID must be robust in the presence of bounce and leakage.

• Ground first, signal later: prevents undefined reference and spurious interrupts.
• Segmented contact: power pins should settle earlier than low-speed I/O.
• Dedicated detect/ID: avoid mixing detect with noisy power pins.

Contact bounce + hot-plug sequencing

Debounce is not a single delay. It is a gating sequence: detect stable → power enable (soft) → voltage-valid window → I/O release → functional attach.

• Too-early I/O: I²C/IRQ glitches occur while the dock rail is still ramping.
• Attach storm: bounce creates repeated attaches; each may restart the dock MCU and flood events.
• Undervoltage behavior: partial power can latch wrong states unless I/O is held in a safe mode.

Contact resistance drift (wear, contamination, corrosion)

Drift converts into failures via Vdock droop (brownout) and slower edges (timing/threshold errors). The risk increases with load steps (e.g., keyboard backlight or scan bursts).

• Evidence: Vdock_min under a known load step; repeated over lifecycle for trending.
• Edge evidence: rising/falling edges stretch; thresholds are crossed late or multiple times.

Exposed-pin ESD (TVS placement + return path)

The dock boundary must close the ESD current loop locally. A TVS that is far from the entry turns the trace into an antenna and forces current through internal references (exactly what breaks touch/pen stability).

• Clamp near entry: place TVS arrays at the pogo boundary, not deep inside.
• Short return: route ESD current to a defined return path without crossing sensitive reference zones.
• Boundary first: treat the dock as a perimeter; do not rely on “more filtering” inside.

Mechanism 1 — Hall sensor + magnet pattern

Provides a presence and coarse pose signal (approach, alignment, polarity sequence). Best used as an early-stage indicator, not a final “ready” proof.

• Good evidence: stable field magnitude/polarity consistent with an intended docking position.
• Common misread: threshold jitter from ground bounce or supply noise in always-on rails. Shows up as rapid toggling.

Mechanism 2 — ID resistor / encoding

Encodes keyboard model/region as an analog level. Sampling must occur after the dock rail is valid and the reference is stable.

• Good evidence: ID voltage falls into a defined bin with margin and remains stable across a debounce window.
• Common misread: sampling during Vdock ramp or during bounce; ground shift moves the level across bins.

Mechanism 3 — Power-pin voltage window + current fingerprint

Proves electrical validity: no short, no severe contact resistance, and no uncontrolled inrush. This is the most direct “is it usable” proof.

• Good evidence: Vdock rises into the window and holds for X ms; droop under a known load step stays above the minimum.
• Common misread: relying on a single instantaneous threshold without hold-time validation.

Mechanism 4 — Simplified confirmation (I²C presence + IRQ)

Confirms that digital communication is possible, but only after controlled power enable and I/O release. I/O should be gated during ramp and bounce.

• Good evidence: I²C address presence success-rate remains high while Vdock stays inside the window.
• Common misread: allowing I²C probing during ramp; glitches generate false interrupts or bus lock.

Where mutual-cap is fragile: scan window + reference

The controller drives Tx at a defined band and samples Rx within a time window. Any reference shift inside the window becomes measurement error.

• Look for: raw data baseline shift and increased standard deviation under charger/dock states.
• Risk: narrowband interference near scan cadence can create beat patterns that look like slow drift.

Harness & routing: guard, ground fence, and gaps

Flexible cables and long routes behave like antennas for common-mode noise. Discontinuous return paths (stitch gaps, splits, via slots) force coupling into the sensing reference.

• Look for: edge-only issues and position-dependent failures (near openings or connector transitions).
• Risk: return detours increase loop area and convert switching ripple into reference movement.

Symptom mapping (hardware-centric)

The same root cause appears differently depending on state and coupling path. Always compare to the Tablet-only baseline.

• Ghost touch: raw noise rises or a periodic interferer aligns with the scan window.
• Edge drift: shielding/return discontinuity near bezel edges and connector zones.
• Worse while charging: VBUS switching ripple and shield/return injection moving the reference.

Evidence: raw data + spectrum peaks + beat behavior

Hardware issues are identifiable by measurement signatures: baseline noise, narrowband peaks, and slow beat patterns when interference and scan cadence interact.

• Raw data noise: compare STD/peak-to-peak across states (Tablet-only vs Charger-attached).
• Spectrum peak: stable peaks indicate coupling from switching rails or ground bounce harmonics.
• Beat pattern: slow drift can be an aliasing result, not a “random algorithm issue”.

Hard constraints: band, bandwidth, and dynamic range

Pen sensing operates in a defined band with a fixed front-end bandwidth and a limited dynamic range. Coexistence with touch requires the pen AFE to survive periodic excitation and harmonics without saturating.

• Band placement: avoid overlap with dominant interference bands and system switching harmonics.
• Anti-saturation: protect LNA/PGA input so touch coupling or rail noise cannot pin the signal.
• AGC behavior: gain changes that are too slow or too aggressive can amplify jitter.

Coexistence with touch (interface-level prerequisites only)

Palm rejection and pen/touch separation require timing alignment and stable thresholds. The requirement is evidence-driven: consistent timestamps or frame markers, plus a noise floor that does not drift with power state.

• Sync evidence: consistent frame markers (timestamp/counter/IRQ) that do not jitter during rail events.
• Noise gate evidence: threshold margin stays stable across Tablet-only and Charger-attached states.
• Sensor assist (interface only): presence/approach signals should not spuriously toggle during switching events.

Wake-on-pen: low-power front-end and IRQ integrity

Pen-approach wake is a low-power detection path that must remain stable on always-on rails. False wakes are a strong clue of reference movement or common-mode injection.

• Look for: wake IRQ count spikes coincident with cable insertion, charger changes, or load steps.
• Risk: insufficient hysteresis or unstable reference in the always-on detection path.

Measurable evidence: jitter, drop points, and spectral peaks

Electrical issues leave signatures. Quantify the symptom, then correlate to interference and rail behavior.

• Pen jitter: point-to-point variance and high-frequency wobble that increases under charging.
• Drop points: missing samples or gaps that repeat at a cadence (often indicates periodic interference).
• Spectrum peaks: stable narrowband peaks that align with switching frequency or its harmonics.

Typical roles in the path (what each block proves)

The path is a chain of roles: input protection, reverse blocking, controlled inrush, source selection, conversion, and rail gating. Each block can be validated by measuring its input/output and timing.

• Input clamp: reduces cable ESD/transient stress near the connector.
• Power-mux / reverse blocking: prevents back-feed and source contention.
• Inrush control / eFuse: limits surge and protects against shorts and contact events.
• Buck/boost: creates stable system rails; switching signatures must not shift references.
• Load switches: sequence rails and isolate sensitive domains during transitions.

Multi-source contention: VBUS vs battery vs dock

Any two sources can fight unless reverse current is blocked and priority is defined. Contention typically shows up as rail oscillation and repeated protection activity.

• Back-feed evidence: a “disconnected” source rises unexpectedly when another source is present.
• Fight evidence: mux output ripple increases during insertion; rails show repeated dips and recoveries.
• Dock path note: if dock supply exists, it must be isolated and current-limited like an external source.

Hot-plug transient: overshoot, undershoot, inrush, droop

Plug-in is a waveform event. Overshoot/undershoot and inrush shape the mux output and can momentarily pull rails below reset thresholds.

• VBUS waveform: capture at the connector; identify overshoot/undershoot and ringing.
• Mux output: verify inrush-limited ramp and absence of oscillation.
• Rail minimum: record Vmin and recovery time; compare to reset/brownout thresholds.

Why touch/pen drift appears when charging (hardware view)

Adapter noise and switching ripple can inject common-mode energy into return paths and shift sensitive references inside scan windows. This is visible as narrowband peaks and state-dependent baseline shifts.

• State dependency: drift increases only in Charger-attached state.
• Spectral signature: stable peaks align with power switching frequency or harmonics.
• Reference movement: sensitive rails (AFE bias/ADC reference/AO ground) show correlated perturbations.

Detach wake scenarios (bind each scenario to a source)

The detachable boundary creates repeatable “wake intents” that can be expressed as a source map rather than vague policy.

• Cover / close / open: Hall + magnet pattern window (stable threshold margin required)
• Keyboard approach / attach: Hall proximity or attach-condition edge (avoid bounce-driven re-attach events)
• Pick-up / motion wake: IMU wake-on-motion (threshold + time window; suppress charger-induced vibration noise)
• Ambient change: ALS threshold crossing (rate limit; avoid flicker-driven wake storms)

Always-on rails and IRQ integrity (who must stay alive)

Always-on (AO) is not a label; it is a defined set of rails and signals that must remain stable in sleep. Leakage and false interrupts typically originate from this domain.

• AO rail scope: sensor supply, IRQ pull network, minimal logic for wake gating.
• Leakage suspects: overly strong pull-ups, level translation left active, floating IRQ inputs.
• IRQ stability: ensure clean edges; avoid marginal thresholds that flip during ground movement.

Suppress false wakes: threshold + hysteresis + debounce

False wakes are most often a threshold and return-path problem, not a sensor “randomness” problem. Make the wake decision robust with explicit margins and time-domain filtering.

• Threshold margin: define windows for Hall/ALS and motion thresholds for IMU.
• Hysteresis: keep decisions from chattering under small field or light fluctuations.
• Debounce window: require stability for N ms or N samples before generating an attach/wake event.

Two-stage confirm (low-power trigger, high-confidence verify)

Detach products benefit from a two-step approach: AO triggers a wake, then the main domain performs a short verification before committing to a full-power state.

• Stage 1: AO interrupt fires with minimal rail and logic.
• Stage 2: read a stable status snapshot (sensor state / event flags) and reject transient triggers.
• Evidence: reduced wake storms without raising AO leakage.

Sensor hub vs direct MCU (power, IRQ, and bus activity only)

The trade is measurable. A hub can reduce main-domain wake frequency, but it may increase always-on current and bus activity.

• Hub benefit: aggregation of wake events; fewer main-domain wake cycles.
• Hub cost: additional always-on device; potential increase in AO baseline current.
• Bus load evidence: compare I²C/SPI activity duty cycle during sleep across scenarios.

Evidence loop: sleep current breakdown + false-wake counters + timestamps

Close the loop with three traces. If any wake source is unstable, the signature appears immediately in counters and current.

• Sleep current breakdown: AO rail, sensors, pull networks, level shifting.
• False-wake counters: per-source counts (Hall/IMU/ALS/attach) rather than a single total.
• Timestamp alignment: correlate spikes with charger insertion, dock attach, or rail transitions.

USB-C ESD loop (connector shell and VBUS)

USB-C brings both direct strike points and cable-borne common-mode noise. Clamp effectiveness is dominated by loop inductance, not only clamp voltage.

• Place clamp near the port: reduce the distance before energy is diverted.
• Choose the return reference deliberately: chassis/shield return can reduce digital ground disturbance when available.
• Measure the loop: if resets occur despite a clamp, suspect a long or shared return path crossing sensitive references.

Pogo pin exposure (detachable-specific fragility)

Pogo pins add variable contact resistance, intermittent connection, and direct exposure. The transient can enter deeper into the board before being clamped if the return path is not local.

• Local clamp array: protect at the interface boundary, not after a long internal trace.
• Return discipline: keep the transient loop local; avoid routing the return through touch/pen reference neighborhoods.
• Evidence hook: zap at pogo area often correlates with keyboard dropouts, touch freeze, or pen loss events.

Keep-out zones (protect sensitive references)

Sensitive references (digitizer reference, AFE bias, ADC reference) should not share return paths with high di/dt currents from switching nodes or interface strike loops.

• Switching nodes: minimize overlap between high di/dt loops and sensitive reference paths.
• Magnetic interface region: keep transient paths from crossing under digitizer reference routing.
• Antenna areas (no RF deep dive): do not route the strike loop across openings near edge radiators.

Common-mode noise → touch/pen noise (path view)

Cable/charger common-mode energy can convert into reference movement through return paths. The signature is state-dependent drift and narrowband peaks.

• State dependency: touch/pen noise rises primarily when a charger or cable is attached.
• Conversion path: common-mode energy couples into ground and reference, shifting the digitizer baseline.
• Evidence: changing adapters moves peak amplitude or frequency in raw noise spectra.

Evidence map: zap point → symptom → fix

Return-path issues are best proven with a zap-point map and before/after comparisons, not only pass/fail statements.

• Zap points: USB-C shell, VBUS, pogo pins, enclosure edges.
• Symptoms: reset cause, touch loss count, pen dropout count, wake storms.
• Before/after: compare noise floor and narrowband peaks after return-path correction.