SPRZ488G March   2022  – September 2026 AM2631 , AM2631-Q1 , AM2632 , AM2632-Q1 , AM2634 , AM2634-Q1

 

  1.   1
  2. 1Usage Notes and Advisories Matrices
  3. 2Silicon Usage Notes
    1.     i2324
    2.     i2348
    3.     i2364
    4.     i2508
  4. 3Silicon Advisories
    1.     i2310
    2.     i2311
    3.     i2313
    4.     i2329
    5.     i2345
    6.     i2346
    7.     i2347
    8.     i2349
    9.     i2350
    10.     i2352
    11.     i2353
    12.     i2354
    13.     i2355
    14.     i2356
    15.     i2357
    16.     i2358
    17.     i2359
    18.     i2374
    19.     i2375
    20.     i2386
    21.     i2392
    22.     i2393
    23.     i2394
    24.     i2395
    25.     i2401
    26.     i2402
    27.     i2403
    28.     i2404
    29.     i2405
    30.     i2427
    31.     i2428
    32.     i2433
    33.     i2438
    34.     i2439
    35.     i2488
    36.     i2499
    37.     i2500
    38.     i2502
    39.     i2503
  5.   Trademarks
  6. 4Revision History

i2499

Incorrect data returned to master on single error detection during burst read

Details:

The RAM controller experiences a timing issue when a single-bit error is detected during an ongoing multi-beat burst read operation. While the ECC module in the RAM controller correctly corrects the error in memory, a handshake timing problem between the ECC module and RAM Controller Bus protocol Logic leads to data misalignment, resulting in incorrect data being returned to the requesting master (R5F’s,DMA,PRU-ICSS,CPDMA, HSM-M4) for subsequent read beats.

  • When a single-bit error is detected during a burst read, the ECC module inserts a BUSY cycle in between.
  • In this BUSY cycle, the next read from memory is halted and a corrected write back is performed to fix the error in RAM.
  • The BUSY signal generated during this process is improperly handled by the Bus Protocol logic.
  • Upon resuming the burst read, the data pipeline becomes misaligned.

Because of this:

  • Incorrect data is returned to the master for reads following error correction.
  • May cause program execution failures and abort exceptions.

This issue affects only L2 and MBOX RAM read operations.

The physical RAM contents remain properly corrected (no memory corruption) and write operations function normally. Below is the high level block diagram.

Workaround(s):

Based on system requirements and safety needs, consider the following solutions:

Option 1: ECC Disabling

For non-safety applications only: Turn off Error Correction Code (ECC) functionality completely.

Option 2: Conservative Error Handling

Configure the system to treat all Single-bit Error Correction (SEC) events as Double Bit Error Detection (DED) events.

This means implementing detection-only behavior without attempting correction, even for correctable single-bit errors.

Option 3: Preventive Memory Scrubbing

If the above options aren't viable, implement periodic memory scrubbing to proactively manage potential errors:

Systematically read memory locations to detect and correct single-bit errors before the application accesses them.

Implementation Methods:

  • SRAM Scrubbing via DMA (Direct Memory Access)

    Refer to SDK reference example for DMA Scrubber

  • SRAM Scrubbing via ICSS (Industrial Communication Subsystem)

    Refer to SDK reference example for PRU Scrubber

Important Limitation: While the scrubbing reduces the probability of encountering errors during normal operation, it cannot eliminate completely the possibility of error occurrence.