DDR MemBIST
be determined from the sequence numbering this is a 10N
pattern.
11.3.5
Error Reporting and Control
Information about failures detected by MemBIST is reported in a variety of ways.
These are summarized here and detailed in the following sections. MBCSR:pf set
during a MemBIST operation indicates that a failure was detected. MBDATA can log the
addresses of up to 5 failures. When the MemBIST operation is utilizing a fixed data
pattern selected using MBCSR:dtype = 0x0, MBDATA also contains a failure data bit
location accumulator, which logs any data bit that has seen a failure during the
operation. With other dtypes, the failure data bit location accumulator may replace the
address logs. MB_ERR_DATA logs the first four 144-bit data words which MemBIST
detects as containing an error after the Memory Test Failure Address Pointer Register
(MBFADDRPTR) has counted down to zero. MBFADDRPTR makes it possible to extract
the failure data information for all repeatable failures which occur during execution of a
MemBIST operation.
How the available logs in are filled depends partly upon the MemBIST configuration and
partly upon how the errors detected by MemBIST occur. For example, when MemBIST
is set to halt on error and a failure is detected, MemBIST does not advance to the next
address. However, due to the architecture of the memory system, there may be more
than one error to log from the single address that was being checked when the failure
was detected. At burst length of 4, for each address there are four 72-bit bus-widths of
data to be checked. Data is logged in 144-bit quantities. If a bit in the DRAM bus was
stuck, this might result in an error being detected in each of the four 72-bit bus-widths
of data. Even though MemBIST is set to halt on error, this would result in two 144-bit
data error logs, each with a corresponding failure address log. Operating at BL=8 could
result in up to four 144-bit data error logs being written with their corresponding
address logs.
When MemBIST is not set to halt on error, logs will fill as errors occur. The logs may
contain information from several unrelated failing addresses. The first four failures to
be detected will cause the first four 144-bit failure data chunks to be logged along with
their corresponding addresses. The address of the fifth error to occur will also be
logged, although there is no location available to log the failure data. The sixth and
succeeding errors will not be logged. During execution of a MemBIST operation,
MemBIST will not overwrite a log entry that has already been filled during that
operation to record a more recent failure.
11.3.5.1
Control Register Pass/Fail Bit and Halt On Error
The MemBIST pass/fail bit, MBCSR:pf, is always cleared when MemBIST execution
starts. If a failure is ever detected, this bit is always set immediately. If MBCSR:halt
(halt on detection of read-compare error) is set, and MBCSR:pf is then set due to
detection of an error, MemBIST execution always completes the accesses to the current
address and then halts without proceeding to the next address. No other MemBIST
controls, including a non-zero value in MBFADDRPTR, modifies the setting of the failure
bit or the halting on failure detection as described.
11.3.5.2
Failure Address Logging
The addresses of the first 5 detected failures can be logged by MemBIST. The address
logs (located in MBDATA) are always cleared, and the logging pointer is always reset to
start logging with the first log, each time that MemBIST starts execution of a new
operation.
128
Intel® 6400/6402 Advanced Memory Buffer Datasheet
INTEL [ INTEL ]
