Executive answer
Four decisions behind a debounce requirement
- Debounce is a state-qualification rule at a named receiver, not one universal time assigned to every membrane switch.
- Press and release need explicit acceptance criteria; hold and repeat timers need a defined starting event.
- Record the raw input, qualified state and application event separately so a supplier change does not hide a controller change.
Current guidance context
What OEM engineers and sourcing teams should decide first
A membrane contact can change between open and closed more than once while a finger applies or removes force. The equipment should still produce the intended logical event once. That outcome depends on the panel, interconnect, receiver circuit, scan schedule and firmware. A drawing note that says debounce 20 ms without naming the observed node or rule does not tell a supplier what must pass.
Begin with the application event. Decide which deliberate presses must be accepted, which short disturbances must be ignored, how soon the command may appear and when release must be recognized. Then map those decisions backward to the raw electrical observation and the controlled build. This turns an implementation setting into a reviewable interface requirement.
Separate contact closure, qualified state and application event
The supplied membrane circuit usually presents an open or closed path at identified tail contacts. The host input converts that path into a voltage or logic observation. A qualification rule then decides whether the observation has remained credible long enough to change the accepted key state. Finally, application logic can convert that state change into a command, hold indication or repeat sequence.
Keep all three layers in the acceptance record. A continuity tester can show that a contact closes and releases under its test circuit, but it does not prove that the target controller samples the same waveform or produces one event. Conversely, a clean application log can hide a marginal raw waveform if filtering happens to suppress it under one configuration.
The receiver also matters electrically. Pull resistance, input leakage, thresholds, capacitance, cable routing and scan timing can change the waveform presented to software. Link the event requirement to the separate input-voltage and resistance budget instead of using debounce to excuse an undefined electrical margin.
Name the qualification rule, not only a time value
Analog Devices documents one concrete implementation in the MAX6816–MAX6818 data sheet: the input is sampled and the output changes after the candidate state remains stable for a specified period. The device treats opening and closing bounce. Its documented 40 ms behavior belongs to that component; it is not a generic requirement for an HYR membrane switch or the correct response time for every product.
Microchip demonstrates a different hardware approach using Timer2 and configurable logic. The first activation starts a timer, while subsequent bounce is ignored until the configured count is reached. That example shows why the same label, debounce, can hide a different state machine. It does not establish a universally safe count or prove that a specific HYR assembly uses that circuit.
For an RFQ, describe the actual rule in plain language or a small state diagram. State whether a candidate change restarts a timer, whether consecutive equal samples are counted, whether press and release use the same setting, and whether an interrupt, sleep mode or scan scheduler changes the sampling interval. A millisecond value without this context is incomplete.
Worked example: one raw trace, two valid but different decisions
This is a hypothetical HYR editorial example, not a customer case, a measured contact trace or recommended firmware. Let 0 mean released and 1 mean pressed. Samples occur every 5 ms. Rule C accepts a change after three consecutive samples match the candidate state. Rule W accepts a change after the candidate state remains unchanged for 20 ms, measured from its first sample.
For the closing sequence 0, 1, 0, 1, 1, 1, 1, 1 at times 0 through 35 ms, both rules reject the early alternation. Rule C accepts pressed at 25 ms because the samples at 15, 20 and 25 ms are all 1. Rule W accepts pressed at 35 ms because the uninterrupted candidate began at 15 ms. Neither answer is inherently correct without a response requirement.
The mirrored release sequence 1, 0, 1, 0, 0, 0, 0, 0 produces the same 25 ms versus 35 ms decision. A two-sample high pulse in an otherwise released trace produces no press under either rule. That may be intentional noise rejection, or it may reject a deliberately short action. The equipment owner must define which pulse widths and user actions matter.
Use the table as a requirement prompt, not a parameter recommendation. Change the interval or rule and the event time changes. The review should compare the proposed implementation with the maximum allowed recognition delay, the minimum deliberate press and the behavior expected when the key releases.
Start hold, repeat and double-press timing from a named event
A hold timer can start from the first raw edge or from the qualified press. Those choices differ by the qualification delay and can diverge further when the raw input alternates. The same question applies to key repeat, double-press windows and audible or visual feedback. Record the chosen starting event rather than assuming every controller uses the same one.
Release deserves equal attention. If the application stops a motion, clears a temporary mode or ends a repeat sequence on release, specify the permitted release-recognition delay and the state after a rejected or interrupted trace. Do not prove the press path and leave the release path to inference.
Also define startup and fault behavior. A controller that powers up while a key is already held might ignore it until release, accept it after qualification or enter a diagnostic state. A stuck contact may need a timeout or inhibit policy. These are equipment decisions, not properties that a passive keypad supplier can determine from artwork alone.
Five acceptance gaps that a clean bench press can miss
Use these scenarios to locate a missing requirement. They are review cases, not reported HYR failures.
- Asymmetric release: the press is qualified correctly, but the release rule or timer is different and leaves the application in a held state longer than intended.
- Timer starts too early: hold or repeat begins on the first raw edge, so a variable qualification interval changes the user-visible duration.
- Sampling mode changes: a low-power or revised scan schedule increases the real interval even though the configured sample count is unchanged.
- Short deliberate input is filtered: a pulse that the application expects is indistinguishable from a disturbance under the selected rule.
- Replacement changes the boundary: a new controller, pull network or firmware revision alters the raw waveform or event logic while the keypad drawing remains unchanged.
Release a testable event contract with the RFQ
Issue one controlled package that links key names and tail contacts to the receiver schematic, polarity, firmware and configuration. For each important input, list the raw sequence or physical action, the qualified press and release states, the expected application event and the permitted timing envelope. Identify who records each layer during sample approval.
Use representative assemblies, connectors, mounting support and intended electronics. Include deliberate normal presses, short presses near the acceptance boundary, slow release, startup with a held key and any relevant hold or repeat sequence. Test combinations separately when the keypad uses a matrix; debounce cannot recover state information that the scanner never observes.
For an HYR review, send the controlled keypad and interface requirements through the secure RFQ. Mark unresolved timing or firmware ownership before freezing the tail pinout and sample plan. A useful supplier response can then state what the component inspection covers and what remains with the assembled-equipment owner.