SENICO Electronics

Component sourcing and inventory guides

UCD9090A Server Sequencing: Dependencies and Fault Recovery

Translate UCD9090A sequencing features into rail dependencies, reset ownership, fault logs and repeatable recovery tests for server control boards.

Quick Answer

UCD9090A should be evaluated as the owner of a defined power sequence, not merely as a convenient collection of enable outputs. TI describes a ten-rail sequencer and monitor supporting dependencies, programmable fault responses and configuration through PMBus. The server designer must specify which rails depend on others, what constitutes a valid rail and how the system recovers when a dependency fails.

The family name identifies the functional device discussed here. Procurement still needs the approved ordering suffix and packaging details. This article proposes a design and acceptance method; it does not establish use of UCD9090A on any particular accelerator assembly.

Key Specifications and Decision Points

TI documents a twelve-bit monitoring ADC, sequencing based on time, rail and pin dependencies, and nonvolatile fault logging. These features can support a reproducible startup and a useful diagnostic record, provided the configuration matches the schematic. A monitor input label or graphical sequence alone is insufficient if the physical net assignment is wrong.

Review threshold values in the units used by the actual configuration tool. Include divider tolerances and the load's permitted voltage range when selecting thresholds. Keep the reasons for those choices in the design record so a later operator does not change them simply to eliminate an inconvenient fault indication.

  • Create a rail dictionary covering net name, nominal voltage, divider, monitor input, enable output and the downstream device that imposes the requirement.
  • Draw startup and shutdown dependencies independently. A valid order for enabling supplies does not automatically define a safe order for removing them.
  • Assign ownership for reset, watchdog and system-level fault signals. Identify whether the sequencer or another device has authority in each operating state.

Application and Acceptance Review

Validate the sequence by observing rails and control signals during the same event. A log indicating that an enable was asserted cannot prove that the output rose correctly at the load. Capture enough information to connect the sequencer's decision with the physical response of the regulator and downstream device.

Fault logs should preserve context. Associate the record with a board identity, configuration revision and event time from the test system. If the device's own record has limited depth or different timing semantics, document that limitation instead of presenting the log as a complete chronological history.

A production configuration package should include the released image, checksum, programming tool version and a readback procedure. Test startup after complete power removal to confirm that the stored configuration, rather than a temporary bench setting, controls the board. Repeat the relevant acceptance cases whenever dependencies or threshold settings change.

  • Test startup with the supported combinations of populated and unpopulated assemblies. A missing optional rail should produce the documented response rather than an unexplained permanent wait.
  • Exercise an approved undervoltage or delayed-start condition on one rail. Check which dependent rails stop, which remain active and how the fault is reported.
  • Restore the faulted condition and verify the intended recovery route, including any required operator acknowledgement or full power cycle.

Pricing and Lead Time

The RFQ for the semiconductor should name the selected full ordering code, required quantity and assembly packaging. The manufacturing request for a board should additionally define programming, verification and traceability. Do not assume that buying the correct sequencer family also delivers the buyer's specific dependency graph.

Ask the assembler how programmed configuration identity will be retained through production and repair. A readable revision label in the build record can save substantial diagnostic effort when nominally identical boards behave differently. If programming occurs before assembly, agree how devices will be segregated and identified so the wrong image cannot be installed silently.

For a configuration service, specify the deliverable and acceptance method before comparing prices. Distinguish preparation of the image, programming of individual units and board-level verification. Obtain actual schedule commitments for each required activity. A device feature list or an evaluation-board demonstration does not establish that a supplier has completed the production configuration for the intended server.

Alternatives and Change Control

An alternative sequencer should be compared using the actual dependency graph and fault policy. Rail count is only the first screen. Monitoring accuracy, input range, log behavior, GPIO assignments, configuration storage and management integration can all influence whether the existing requirements remain achievable.

If a design distributes sequencing across several devices, define how their states and faults coordinate. A local fault on one board may need to inhibit another board even when the management processor is unavailable. Test this coordination explicitly rather than assuming that separate working sequences combine into a working system.

Software-controlled sequencing is another possible architecture, but it changes startup dependencies and failure handling. Determine whether the processor responsible for sequencing can boot before the rails it controls and what happens if its software hangs. The final choice should make ownership clear in every state. Preserve that ownership model with the released schematic and configuration so future maintenance can diagnose the system without reconstructing its logic from scattered register settings.

FAQ

Can I define shutdown by reversing the startup list?

Use the downstream devices and regulator behavior to establish shutdown requirements independently. Stored energy, back-feeding and reset requirements can make a simple reversed list unsuitable for a particular design.

Why test an optional board being absent?

Population changes can alter dependency satisfaction and monitored voltages. Qualification should show the intended behavior for supported configurations, including which absent rails are acceptable and how the sequencer identifies them.

What should accompany a programmed UCD9090A assembly?

Retain the board revision, released configuration identifier, programming and readback evidence, and relevant sequence test results. These records connect the physical circuit with the logic controlling its operation.

Discuss your requirements with SENICO and request a quotation

Sources