Establish a comparable read baseline
Start with a repeatable difference between two read paths on the same W25Q128JVSIQ. Choose existing content containing varied bytes. An all-zero test area cannot expose this symptom. Record the board revision, host ordering code, clock, firmware version, and initialization sequence.
Use one flash offset and transfer length. For the mapped access, translate that offset through the host's documented memory window. A CPU address and a flash command address are not interchangeable. Save the actual bytes, not just an API success status.
| Record item | Indirect read | Mapped read | Evidence to retain |
|---|---|---|---|
| Address | Flash offset | Host window plus that offset | Address calculation |
| Content | Known bytes and returned bytes | Returned bytes at the same length | Byte comparison |
| Read conditions | Opcode and clock | Opcode and clock | Configuration and bus capture |
| Outcome | Status and actual payload | Actual payload and errors | Capture before and after changes |
This record defines the failing transition. If basic identification also fails, use the separate all-FF JEDEC ID checks first. The following steps assume that indirect reads reproduce the expected content.
Check the rail transition and the issued read frame
Capture the transition and first mapped access from the baseline test. Probe VCC at the flash pins, with /CS and clock as timing references. Check the supply against the operating range for the actual frequency. A steady regulator reading elsewhere does not establish the voltage during this event.
Next, decode the transaction actually issued after mapping starts. Compare instruction width, address width and value, mode bits, dummy clocks, and data width. A driver configuration screenshot cannot establish what reached the device.
The Winbond W25Q128JV datasheet Revision M specifies different frames for these commands. For 6Bh, both the instruction and 24-bit address use one line. Eight dummy clocks precede four-line data output. In SPI mode, EBh sends its instruction on one line. Its address, mode bits and data use four lines, with four dummy clocks before data. These phase settings are not interchangeable.
For SIQ, the IQ ordering option has QE factory-fixed at 1. Do not make a generic QE write the assumed repair. Do not import an IM variant's QPI sequence into this configuration.
If the rail and frame are valid, trace the host window, receiving path, and cache or memory attributes. Follow the exact host documentation and errata. ST's external-memory interface guidance also warns that caching can mask faults. Establish a controlled comparison without assuming universal register settings.
Correct the path supported by the evidence
Choose the next action from the failing stage. Test one evidence-based hypothesis at a time. Keep the original payload, address, and capture trigger so the comparison remains useful.
| Observation | Supported assessment | Action or next check | Retest conditions |
|---|---|---|---|
| VCC falls as mapping starts | Supply conditions need investigation | Check the switch, source impedance, decoupling, and load | Same transition and probe point |
| Stable rail but wrong command phases | Observed frame disagrees with the selected instruction | Correct the specific phase mismatch | Same offset and known bytes |
| Correct payload on IO0–IO3 but CPU sees zeros | Investigate host reception and access | Check window, sampling, memory attributes, and cache | Controlled read with matching capture |
| No /CS during an access expected to reach the bus | Confirm whether the request reaches the controller | Check address, controller state, and cache effects | Exclude a cache hit before judging /CS |
A lower clock that restores data narrows the investigation; it does not prove which timing parameter failed. Likewise, a supply dip does not by itself identify excessive flash current. Preserve the evidence and investigate the corresponding circuit or timing condition.
Verify the outcome against the original baseline
A repair is supported when both read paths repeatedly return the expected bytes under the original failure conditions. Update the first table after each change. Save the results from the transition that originally triggered the failure.
Retain the changed setting or circuit item, firmware version, and matching captures. If checking additional clocks or power cycles, record their conditions separately. A pass at one operating point does not qualify every supported frequency.
For a component review, use the Winbond part directory to locate the exact ordering code. Keep device selection separate from board-level acceptance. Conclude with the verified scope, unresolved discrepancy, and next measurement needed.
For a W25Q128JVSIQ project review, prepare the host part number, supply conditions, read commands, and failing comparison. Check candidate parts against their ordering options and the validation your board requires. For a SUNTEC quotation, include the complete part number, quantity, and required delivery date.
Prepared by the SUNTEC Technical Team at Suzhou SUNTEC Electronics Co., Ltd.