Method and apparatus for memory self testing

[0001] 1. Field of the Invention

[0002] This invention relates to the field of data processing systems. More particularly, this invention relates to the self-testing of memories within data processing systems to detect memory defects.

[0003] 2. Description of the Prior Art

[0004] It is known to provide data processing systems incorporating memories with self-testing mechanisms, sometimes termed built-in self-test (BIST), such that when the memory circuits have been fabricated they may conduct a self-test or series of self-tests to determine if there are any memory defects present that would indicate that the circuit should be rejected. It is important that memory circuits should be highly reliable and that defective circuits should be effectively identified. As memory sizes increase, the proportion of the die being taken up by the memory is also increasing and, furthermore, the high transistor density within memory means that the defect density of these memories is becoming increasingly significant.

[0005] The type of defect that can arise may vary significantly from design to design fabrication processes applied to the same design. For this reason, self test methodologies are normally selected and tuned to a particular design and fabrication technique by the manufacturer concerned. A manufacturer will typically have their own preferred test methodology that experience has told them is well suited to their particular environment to identify defective memories. The test methodology adopted by one manufacturer may well not work well for another manufacturer.

[0006] It will be appreciated that self-test circuits whilst necessary to ensure a memory has been correctly fabricated, do represent a circuit area overhead that should be desirably reduced. The self-test circuits are typically used when an integrated circuit is manufactured but thereafter are not used during the normal operation of the integrated circuit.

[0007] Viewed from one aspect the present invention provides apparatus for processing data, said apparatus comprising:

[0008] at least one memory having a plurality of memory storage locations associated with respective memory addresses;

[0009] a self-test controller operable to control self-test of said at least one memory including generating physical memory address signals; and

[0010] a mapping circuit operable to map said physical memory address signals generated by said self-test controller to corresponding logical address signals for use by said at least one memory to perform a memory test based upon a physical position of memory storage locations.

[0011] Providing a self-test controller that generates physical memory address signals and then using a mapping circuit to map these to logical address signals facilitates the reuse of the self-test controller. Different memories may have very different mappings between the physical memory locations (topological addresses) and the logical addresses of those locations. However, in terms of memory test methodology the physical address is much more significant than the logical address and accordingly a generic self-test controller can provide good test methodology coverage applicable to a wide variety of memory designs by generating physical memory address signals and relying upon a relatively small and easy to design mapping circuit to map those physical memory address signals to corresponding logical memory address signals as required to drive a particular memory device during test. The reuse of a self-test controller in this way is highly beneficial. In the context of system-on-chip designs that make increasing use of circuit elements designed by different sources, the mapping circuit enables a generic controller to be used more readily.

[0012] A particularly convenient way in which physical memory address signals may be represented is to use column addresses and row addresses. The generation of appropriate address sequences for addresses expressed in this way is more straight forward than with addressing structures that more abstractly represent the physical layout of the memory device.

[0013] It will be appreciated that the mapping circuit operates so as to preserve the relative physical position of memory addresses whether they are expressed in terms of physical memory address signals or logical memory address signals.

[0014] A common example of different memory devices for which different mappings may be desired is where the memory device can be one of a custom memory or a synthesisable memory.

[0015] The mapping circuit may be conveniently provided within an interface circuit disposed between a self-test controller and a memory circuit. Such an interface circuit can also adjust other signal values and signal timings as required to adapt the self-test controller to a particular memory circuit. The concentration of the customisation needed for different memory devices within such an interface circuit strongly facilitates the design reuse advantages of the present technique.

[0016] The present technique is particularly well suited to embodiments in which a processor core and memories are fabricated together on a single integrated circuit. The wide variety of possible combinations that such systems provide make the reuse of a self-test controller design strongly advantageous compared with having to custom design a self-test controller for each particular circumstance.

[0017] The invention is also well suited to systems in which multiple memories are provided on a single integrated circuit. The multiple memories may have many different characteristics. As an example, a cache memory could be a custom device whereas an on-chip RAM (such as SRAM) may be synthesized. In such a situation, the two different memory devices may have greatly differing mappings between the physical position of a memory location and its logical address. In such circumstances, it is desirable to use a single self-test controller to test all of the memories and the provision of appropriate mapping circuits for each memory device allows this option.

[0018] Viewed from another aspect the present invention provides a method of testing a memory having a plurality of memory storage locations associated with respective memory addresses, said method comprising the steps of:

[0019] generating physical memory address signals; and

[0020] mapping said physical memory address signals to corresponding logical address signals for use by said at least one memory to perform a memory test based upon a physical position of memory storage locations.

[0021] The above, and other objects, features and advantages of this invention will be apparent from the following detailed description of illustrative embodiments which is to be read in connection with the accompanying drawings.

[0022] FIG. 1 schematically illustrates an integrated circuit incorporating a processor core, multiple memories and a self-test controller.

[0023] FIG. 2 schematically illustrates a self-test controller, an interface circuit and a memory in more detail;

[0024] FIG. 3 is a flow diagram schematically illustrating the use of the self-test controller of FIG. 2;

[0025] FIG. 4 schematically illustrates a mapping of physical memory address signals to logical memory address signals; and

[0026] FIG. 5 illustrates in more detail a particular example of an address signal mapping that may be required.

[0027] FIG. 1 shows an integrated circuit 2 incorporating a processor core 4, a cache memory 6 and an RAM memory 8 (in this example SDRAM). It will be appreciated that such an integrated circuit 2 will typically include many more circuit elements, but these have been omitted from this drawing for the sake of simplicity and clarity.

[0028] A self-test controller 10 is provided on the integrated circuit 2 and communicates with each of the cache memory 6 and the SDRAM memory 8 by a respective dispatch circuit 12, 14. These dispatch circuits act as interface circuits between the self-test controller 10 and the respective memories. Thus, generic memory test driving signals produced by the self-test controller can be re-mapped to logical addresses and adjusted in value and timing to the requirements of the different memories without a need to change the fundamental design of the self-test controller 10.

[0029] It will also be seen that the self-test controller 10 includes a self-test interface 16 provided in the form of multiple external signal pins of the integrated circuit 2. In this way, when the integrated circuit 2 has been fabricated, it may be plugged into an appropriate test controller that initiates the self-test action of the self-test controller 10 and responds to any detected memory defects by either rejecting the integrated circuit 2 as a whole or applying appropriate repairs in circuits where this is possible.

[0030] FIG. 2 schematically illustrates the relationship between the test controller 10 and a memory device 18 in more detail. The self-test controller 10 includes a self-test instruction register 20 into which self-test instructions can be loaded via the self-test interface 16 using serial shifting techniques. Once the self-test instruction is in place, self-test action may be triggered by assertion of an appropriate externally applied signal, such as MBIST_RUN. Within the self-test controller 10 a state machine 22 is responsive to a pattern selection portion of the self-test instruction which serves to select which physical memory address sequence will be generated by the self-controller 10 for application to the memory device 18. As examples, the patterns selected may serve to perform the following actions:

[0031] write a specified data word into all entries;

[0032] read a data word from all entries;

[0033] write a data word or inverse data word in a checkerboard pattern into all entries;

[0034] read a data word or inverse data word in a checkerboard pattern from all entries;

[0035] perform a march C test (this type of test is well known in the memory test field and will not be described herein further);

[0036] perform a read operation and then a write operation using an increment decrement word line fast march;

[0037] perform a read operation and then a write operation using an increment decrement bit line fast march;

[0038] perform a read operation, a write operation and a read operation using an increment decrement word line fast march;

[0039] perform a read operation, a write operation and a read operation using an increment decrement bit line fast march;

[0040] perform a bang test which serves to stress test bit line pairs by repeated write operations and read operations;

[0041] perform a predefined suite of tests (such as those mentioned above) to yield a go/no go result; and

[0042] perform a pattern which creates false read data at specific points in order to validate fail detection.

[0043] The data word that may be applied as part of the above described techniques is also supplied as a bit field within the self-test instruction. An engine-control field within the self-test instruction serves to define error reporting options for the self-test controller 10 and self-test system as a whole. An array selection portion of the self-test instruction serves to select which particular memory device is to be tested with the self-test instruction concerned in an environment containing multiple memory devices. It will be appreciated that a self-test instruction could be applied to multiple memory devices in parallel in appropriate circumstances.

[0044] The self-test controller 10 generates physical memory address signals in the form of column address signals and row address signals that are passed to a dispatch controller 24. The data word to be applied in the specified test is also passed to the dispatch controller 24. Various other control signals are also exchanged between the self-test controller 10 and the dispatch controller 24.

[0045] The dispatch controller 24 acts as an interface circuit for applying any necessary signal value adjustments or timing adjustments to the signals generated by the generic self-test controller 10 that are required in order to drive the desired test within the memory device 18. As an example, different latencies of signals may be accommodated by incorporating different lengths of pipeline delay between memory access instructions being received and being serviced. A further significant role of the dispatch controller 24 is to map the physical memory address signals generated by the self-test controller 10 into required logical memory address signals that may drive the memory device 18. In the case where the dispatch controller 24 has been specifically designed for the particular memory device 18, then this mapping may be incorporated into that design. In other embodiments, it may be that the dispatch controller 24 itself is generic to some degree and it is desirable to be able to deal with different mappings that may be necessary. This may be optionally accommodated by latching the physical memory address signals, supplying these to a further separately designed and provided mapping circuit 26, before the desired logical memory address signals are returned and latched within the dispatch controller 24. This separately provided mapping circuit 26 is illustrated within a dotted line within FIG. 2 to demonstrate its optional character. It will be appreciated that the mapping circuit may be separately provided in this way or may be incorporated within the dispatch controller 24. Both of these possibilities are encompassed within the present technique.

[0046] The dispatch controller 24 provides signal values necessary to drive the test within the memory device 18. These signals include a logical address value, a data value, which will typically be an expanded form of the data word as specified within the self-test instruction, and any associated control signals. The dispatch controller 24 will also monitor the response of the memory device 18 to the requested memory access operations and report detected memory defects, such as an incorrect data bit being read back, to the self-test controller 10. The dispatch controller 24 may include data registers that serve to log data specifying any detected errors or other parameters. This data may be read from the dispatch controller 24 to the outside environment using an appropriate self-test instruction scanned into the self-test controller 10 after the testing is complete. This type of operation is particularly useful in design debugging where a fine level of detail regarding detecting errors may be required.

[0047] FIG. 3 is a flow diagram schematically illustrating operation of the embodiments illustrated in FIGS. 1 and 2. At step 28 a self-test instruction is serially loaded into the self-test controller 10. At step 30 control is passed to the self-test controller 10 by the external application of the MBIST_RUN signal illustrated in FIG. 1. At step 32, the state machine 22 within the self-test controller reads the pattern select bits, the engine control bits and the data word bits from the self-test instruction stored within the self-test instruction register 20 and then generates the specified test signals at step 34. It will be appreciated that the self-test instruction will typically specify a long sequence generating many thousands of physical memory address signal values and data values to be applied to the memory device under test to carry out at least part of a desired test methodology.

[0048] After the particular self-test instruction has completed execution, a test at step 36 is made as to whether a pass result or a fail result has been returned. If a fail result has been returned, then processing proceeds to step 38 at which any desired fail action is triggered. If the result was a pass then the processing of the particular self-test instruction terminates directly.

[0049] It will be appreciated that the flow diagram of FIG. 3 illustrates the execution of a single self-test instruction. In practice, a test methodology will typically be built up by a manufacturer or user in the form of a long sequence of such self-test instructions as required to specify a test methodology specially adapted to that particular design and fabrication environment. Thus, multiple self-test instructions will be loaded into the self-test controller 10 and multiple passes through the processing illustrated in FIG. 3 will be made. Data retention, read disturb and iddq testing are supported to be controlling the pattern set.

[0050] It will be appreciated that some users may not wish to custom specify their own test methodology and accordingly may rely upon the default test methodology mentioned above which could be specified using a single self-test instruction and return a single simple go/no go result value.

[0051] FIG. 4 illustrates the receipt of physical memory address signals 40 from the self-test controller 10. These are manipulated by a mapping circuit 42 to generate appropriate logical memory address signals 44 that can drive the memory device under test. In a simple arrangement there may be a one to one mapping of the physical memory address signals to the logical memory address signals. However, in more complicated situations the relationship between physical memory address signals and logical memory address signals may be more complex, but nevertheless readily achievable with a comparatively simple and easy to design mapping circuit 42.

[0052] FIG. 5 illustrates an example of an address signal mapping that may be provided. The memory device under test is formed of two memory arrays each containing 32 rows and 8 columns. The logical address is a 9 -bit value. The most significant logical address bit LA8 is used to control a multiplexer to select between the two arrays of memory. The least significant 3 -bits of the logical address LA2:0 is used to select the column. The remaining logical address bits LA7:3 are used to select the row. This relationship and arrangement is illustrated at the top of FIG. 5.

[0053] The middle portion of FIG. 5 is similar to FIG. 2. This illustrates a macrocell which may be part of a SOC design (and have its own an SOC interface), which also includes a self-test controller 10, a dispatch controller 24 as well as the memory device formed of the two blocks as discussed above. The self-test controller 10 treats the memory device as an array of memory locations having 32 rows as represented by the 5 -bit row address Xaddr. The column of the memory locations is expressed as the Yaddr 4 -bit address irrespective of whether this column is in the lower order block or the higher order block within the physical device. In the example of FIG. 5 the mapping circuit is provided externally of the dispatch controller 26 via a custom circuit attached to the SOC interface. Attaching the custom mapping circuit at this level allows easy alteration to this circuit and design of this circuit. In the example illustrated, the physical memory address signals are latched within the dispatch circuit 24 and then passed out to the separate mapping circuit. Within the mapping circuit the least significant 3 -bits of the Yaddr field form the least significant 3 -bits of the logical address LA2:0. The most significant bit of the Yaddr value forms the most significant bit LA8 of the logical address. The row address 5 -bit value Xaddr forms the bits LA7:3 of the logical address. The logical address LA 8:0 built up in this way is passed back from the mapping circuit into the dispatch controller 24 and again latched prior to being applied to the memory device to perform the desired test.

[0054] A further more detailed description of the above techniques is given as follows:

Contents
1	About this document
1.1	Change history
1.2	References
1.3	Terms and abbreviations
2
2.1	SCOPE
2.2	Summary
2.3	MBIST Architecture
2.3.1	Summary
2.3.2	MBIST User interface (test access to BIST Ctl).
2.4	MBIST Instruction Resister
2.4.1	Array Enables and Flags <:16> (must be in nibbles increments)
2.4.2	DataWord <3:0>
2.4.3	Pattern/Algorithm selection <9:4>
2.4.3.1	Definitions
2.4.3.2	BIST Pattern Specification
2.4.4	Engine Control <15:10>
2.5	MBIST Memory interface
2.5.1	Memory BIST Interface Definitions
2.5.2	MBIST Interface Design Rules
2.6	MBIST Dispatch Units
2.6.1	Dispatch Unit interface
2.6.2	Address Scrambling
2.6.3	Size Variance
2.7	Design Pilot Program

[0055] This specification provides an outline for implementation of memory BIST in both hard and soft core products. This specification will act as a blueprint for all new designs within ARM with the end goal of creating a re-useable and constant memory BIST architecture and memory BIST interface for all customers (ARM internal included).

[0056] The memory BIST architecture will require functionality to exist in both hard and soft cores since it is ARM's desire to place memory test interfaces in non-critical paths. This access can be achieved by accessing write and read paths n pipeline stages away from the physical array. Physical mapping of arrays is required for best array test quality and soft cores will provide an address porting to allow the end user to re-configure address pins to match physical implementation once it is known.

[0057] ARM will provide an optional memory BIST controller in soft core implementations and will provide a custom implementation with hard cores (not optional). It is clear that every end customer has their own set of beliefs when it relates to memory BIST algorithms. It is ARM's goal to provide standard and custom algorithms which test for numerous fault classes and types. While these patterns may not perfectly match every end user's test methodology, they should be sufficiently robust such that the end user is not needing to create a new implementation.

[0058] There are two interfaces receiving special attention in this specification. First is the user interface. The user interface is specified with future expansion in mind. Use from core to core will remain the same, as will the primary instruction register that is loaded by this interface. This provides the same look and feel across the product line. The second interface exists at the memory access points. It is important to have a common interface among various cores to insure that the end user's own BIST solution can be configured for all future ARM cores with a common interface as well as meeting requirements for our internal BIST engine.

[0059] This document does not address specific ATPG test needs such as array bypass or controls but does include a couple of features that are required regardless of ATPG testability design style.

[0060] ARM memory BIST strategy revolves around the ability to run testing with physically mapped arrays in order to leverage pattern effectiveness. Hard cores are always physically mapped in X (rows/distance to/from sense amp) and Y (column or sense amp selection) dimensions. Soft cores will have documented mappings of the memory test interface to the logical address provided at the core to memory interface. It will be a required exercise of the integrator (if they are to realize maximum test effectiveness ) to either:

[0061] 1) perform the physical mapping by scrambling the BIST_dispatch unit for each interface to match intended mapping

[0062] 2) Force physical mapping in compiled rams to match the predefined ARM mapping

[0063] The Memory BIST architecture will be centered around a single memory BIST test engine which distributes control signals to all arrays in the core. It is assumed that one memory BIST implementation will be created per core although it may be possible to integrate an engine across multiple cores. Providing two memory BIST engines per core is allowed but tester to memory BIST controller test interface must be managed and selected by the system integrator. Test signal distribution and routing impact is minimized by the use of a 4 bit dataword bus leaving the memory BIST controller. Dataword expansion and testing will occur at a specialized unit called mBIST_dispatch which acts as glue logic between the memory BIST controller and the memory BIST test interface. This glue logic performs address scrambling (mapping to physical space), specialized pipelining (local to targeted arrays), and dataword comparisons and test fail datalogging. This significantly reduces the excessive routing overhead usually associated with a centralized BIST controller.

[0064] The dispatch unit containing glue logic provides design specific implementations to be accommodated while enabling the memory test interface and BIST controller to be a fixed standard from design to design. End user test interface can be a standard from design to design by having a BIST controller that is off the shelf re-usable in all future ARM cores.

[0065] Memory BIST is launched by configuring the core's test interface to allow access to the memory BIST interface. Test algorithm progress (measured by address expire flags) and pass/fail status are continuously delivered on a 4 bit output only data bus (ported to macrocell). Further datalogging information can be obtained by shift access (using same external BIST interface) of the BIST controller or failing unit's dispatch unit. The amount of datalogging and trace support in the dispatch unit will be design dependent and may be configurable.

[0066] Each memory BIST interface may support multiple arrays. Each of these arrays will be supported as separate arrays in the BIST instruction register and failure log. During test execution, these arrays will also be tested by appending their enable lines to the Yaddr space. A BIST dispatch unit can accommodate mapping of enable lines to higher address bits if so enabled by the instruction register's enable lines. Note that this does not fit the usual definition of serial test or parallel test of these target arrays but the arrays are instead treated as segments of a larger whole array.

[0067] Centralized BIST controller and dispatch units:

[0068] The MBIST test interface is defined by the below table. Of concern is the access mechanism for shift loading of the test configuration. This shift process will be performed on the core clock domain. As cores progress into the future, this may require a pll switching mechanism to allow slow manual shift (due to tester interface limitations) and a switch to high speed pll clocking in order for memory test to occur at speed (if clock rate exceeds the tester cycle or if pad timing limitations suppress maximum clocking, speeds). It may make sense to implement a divide by four shifting clock for Jaguar in order to support future migration paths.

Test Interface to MBIST
Core Port Signal	I/O, Size	Description
MBIST_RUN	In	Memory testing launched
		by this signal
MBIST_SHIFT	In	Enables control register
		data load
MBIST_DIN	In	Provides serial load of control
		registers
SCANMODE	In	Disables overrides at
		dispatch units
MBIST_DOUT	Out, <3:0>	Provides runtime status
		<done, fail, Xexpire, Yexpire>

[0069] The test configuration is loaded into a 32 bit register by asserting MBIST_SHIFT for 32 clocks and delivering LSB first, MSB last. MBIST_RUN MUST be inactive during this register load. Execution of register load terminates when MBIST_SHIFT is de-asserted. Assertion of MBIST_RUN will begin the memory test selected and test termination is signalled by the MBIST_DOUT<3> bit. MBIST_DOUT<1:0> will toggle during test to mark algorithm progress. MBIST_DOUT<2> or the fail bit will be sticky.

[0070] MBIST_SHIFT will be gated by MBIST_RUN to prevent unknown hazards in the engine.

[0071] MBIST_DOUT<0> doubles as the shift data out path of the instruction register whenever (MBIST_SHIFT & MBIST_RUN).

[0072]

[0073] Each tested array in the core will be assigned an enable bit as specified to the TRM document. This implementation is defined in this manner to account for the desire to test all arrays in parallel and to also allow flexibility to test arrays individually should the need arise. One example is the need to test only 50% of the arrays per pass due to maximum power concerns. Another possibility may include initialising different arrays to different background patterns for use in ATPG or in iddq. An array is not defined by the number of memory test interfaces but is defined by the number of individually tested arrays (note that a single memory test interface may have multiple array enables).

[0074] This portion of the instruction register also serves a second purpose. The outside user sees a single failure flag on the external interface. Scanout of this portion of the register, will reveal the P/F status of each array to further identify the failure array.

[0075] The maxYaddr field allows the user to specify the maximum Yaddr (a.k.a. column addr) to be used during test. This is especially useful when core derivatives of with varying cache sizes are employed or when different sized arrays within the core are to be tested. This bist controller architecure does not prevent multiple arrays of different sizes from being tested at the same time. To perform test of such scenarios, the user programs a maximum address size equal to the largest array and the dispatch unit blocks out of range addresses (by gating RE, WE) for the smaller array.

[0076] The following table describes the register settings and resulting address sizes:

MaxYaddr	Size	BistYaddr active bits 11;0
0000	0	000000000000
0001	2	000000000001
0010	4	000000000011
0011	8	000000000111
0100	16	000000001111
0101	32	000000011111
0110	64	000000111111
0111	128	000001111111
1000	256	000011111111
1001	512	000111111111
1010	1 k	001111111111
1011	2 k	011111111111
1100	4 k	111111111111
>1100		Not supported

[0077] Similar to the Y address, the X address field slects the maximum number of addressable rows. Row depths beyond 512 are usually impractical due to bitline RC's but a space of up to 2 k is supported here if the user wishes to map planar features in the topology to the Xaddr space (we recommend that planar or block addresses be mapped on top of Y address space, else bitline stress tests based on Xaddr space may lose effectiveness).

MaxXaddr	Size	BistXaddr active bits 10:0
0000	0	00000000000
0001	2	00000000001
0010	4	00000000011
0011	8	00000000111
0100	16	00000001111
0101	32	00000011111
0110	64	00000111111
0111	128	00001111111
1000	256	00011111111
1001	512	00111111111
1010	1 k	01111111111
1011	2 k	11111111111
11xx		Not supported

[0078] The engine control portion of the register is used to specify special state machine behaviors such as stop on error. An <15:10>0 setting specifies default behavior.

<15:10>	Behavior	Description
00 0000	Default	Test runs to completion
1x xxxx	Real Time Fail bit	Fail bit is real time, sticky fail
		appears at end of test
xx 0001	Stop on fail	Stops on fail with <16
		cycle latency
xx 0010	Stop at end of algorithm if fail	Stops on fail after current
		algorithm N is completed
00 1110	Load dispatch unit instruction	Scan dispatch unit as selected by
		one hot <31:16>
00 1111	Read dispatch unit datalog	Scan dispatch unit as selected by
		one hot <31:16>

[0079] The real time failure bit is provided for lab use should the test/lab personnel wish to monitor soft fail behaviors.

[0080] Stop at end of algorithm pass is provided as a means to allow testing to progress until end of a failing algorithm segment. This may be useful for bitmapping if the dispatch unit provides scannable test access mechanisms for downloading all array entries.

[0081] Scanning of dispatch unit will place selected dispatch unit scan ports between MBIST_DIN and MBIST_DOUT when MBIST_RUN is asserted. Access is disabled once MBIST_RUN is de-asserted. Any operation loaded into dispatch unit will be executed at time of MBIST_RUN de-assertion. (Note: I need to better define handshake to allow maximum efficiency and also proper function as mBIST_run will require warm-up cycles and must allow enough time for the operation to complete, would probably be ok if I delayed internal override releases by appropriate cycle counts).

[0082] Dispatch units are loosely defined at this time and will be built to core's need. They should provide minimum datalogging information, such as first fail address and fail bit. This information should be accessible by the dispatch datalog scan operation. Use of scannable dispatch unit instructions is optional.

[0083] Numerous standard patterns and ARM created bitline stress patterns are provided for use in test. Algorithms are created in pieces to allow for maximum flexibility in unknown final test design. Areas of test supported include go/nogo, iddq, data retention, ATPG, gross function test, and stress testing. The following table describes supported patterns and later paragraphs describe their use. Use of segmented testing can be used to maximize softbin and yield tracking information in the fab and to provide the end test design to best match the fab's memory test own beliefs.

<5:0>
Selec-
tion	Name	N	Description
00000	WriteSolids	1 N	Write specified dataword into all entries
00001	ReadSolids	1 N	Read dataword from all entries
00010	WriteCkbd	1 N	Data or databar selected
			by X<0> {circumflex over ()} y<0>
00011	ReadCkbd	1 N	Data or databar selected
			by X<0> {circumflex over ()} Y<0>
00100	March C	14 N	Increment decrement XY fast
			RWR march
00101	PttnFail	6 N	Tests BIST failure detection
00110	RW_Xmarch	6 N	Increment decrement wordline
			fast march
00111	RW_Ymarch	6 N	Increment decrement bitline fast march
01000	RWR_Xmarch	8 N	Increment decrement wordline
			fast march
01001	RWR_Ymarch	8 N	Increment decrement bitline fast march
01010	Bang	18 N	Increment decrement wordline fast
			bitline stress
11111	Go/No Go	30 N	2*W/RCkbd, RWR Ymarch, X Bang

[0084]

column	>	dimension in array parallel to bitlines (on the
		same SenseAMP)
row	>	dimension in array parallel to wordlines
row fast/Xfast	>	target cell moves along bitlines before moving
		to next column
col fast/Yfast	>	target cell moves across bitline pairs before
		row/wordline
Xfast Increment	>	target cell begins nearest sense amp,
		moves away
Xfast Decrement	>	target cell begins furthest point from sense
		amp, moves closer
Yfast Increment	>	Yaddr space moves from 0 to max, east-west
		relationship
Yfast Decrement	>	Yaddr space moves from max to 0, opposite
		of increment
N	>	Number of addressable entries

[0085] All memory BIST tests are executed against physically mapped address space. This simply means that the least significant Xaddr switches between adjacent rows (LSB1 switches between every 2nd row, and so on). Yaddr space is also physically mapped. This allows for efficient and direct targeting of memory faults with the supplied patterns.

[0086] writeCkbd is performed Xfast. This pattern is 1N, writing only. Data polarity set by xor(Xaddr0, Yaddr0).

[0087] ReadCkbd is performed Xfast. This pattern is 1N, reading only. Data polarity set by xor(Xaddr0, Yaddr0).

[0088] writeSolids is performed Xfast. This pattern is 1N, writing only. Data polaritytrue.

[0089] ReadSolids is performed Xfast. This pattern is 1N, reading only. Data polaritytrue.

[0090] MarchC is performed Xfast with the following sequence:

[0091] 1) write solids (initialize array) incr

[0092] 2) Rdata, Wdatabar, Rdatabar incr

[0093] 3) Rdatabar, Wdata, Rdata incr

[0094] 4) Rdata, Wdatabar, Rdatabar decr

[0095] 5) Rdatabar, Wdata, Rdata decr

[0096] 6) Rdata verify decr

[0097] RWR_Xmarch is performed Xfast with the following sequence

[0098] 1) write solid (initialize array)

[0099] 2) Rdata/Wdatabar/Rdatabar inc

[0100] 3) Rdatabar/Wdata/Rdata decrement

[0101] 4) Rdata solid

[0102] RWR_Ymarch has the same sequence as RWR_Xmarch but is performed Yfast.

[0103] RW_Xmarch is performed Xfast. This 6N pattern has the following sequence

[0104] 1) write solid (initialize array)

[0105] 2) Rdata/Wdatabar inc

[0106] 3) Rdatabar/Wdata decrement

[0107] 4) Rdata solid

[0108] RW_Ymarch is 6N and performed Yfast with the same sequence as RW_Xmarch.

[0109] pttnFail is performed Xfast. Executes a writeSolid pattern followed by a ReadSolid. Fails are injected by reversing data polarity on select addresses during ReadSolid. This pattern is REQUIRED to insure BIST detection logic (at target array) is functional.

[0110] bank, 18N, is performed Xfast, executing consecutive multiple writes and reads on a bitline pair.

[0111] 1) write solid (initialize array)

[0112] 2) read data target, write databar target, 6 * write databar row0, increment to next target Xfast

[0113] 3) 5 * read databar target, write data row 0, read databar target, write data target

[0114] 4) read data (verify array)

[0115] This segment bangs bitline pairs insuring proper equalization after writes. Insufficient equalization/precharge will result in a slow read when opposite data is read from the same bitline pair. Slow reads in self timed caches result in functional failure not found in popular single shot algorithms like March C.

[0116] This segment attempts to walk down a bitcell, write opposite data on that bitline pair, read target cell data. This failure mechanism is less common in 6T ram cells (vs. 4T or DRAM).

[0117] Use of the sacrificial row, also helps detect open decoder faults in the Xaddr space (Yaddr not subject to fault class architecturally because of smaller decode blocks) in absence of graycode pattern sequences.

[0118] This pattern does detect stuck at fault, but its primary intent is to address the memories' analog characteristics.

[0119] go/nogo, 30N is the ARM test suite selected for memory tests should the end user not desire to implement their own test strategy. Selection of this pattern will result in the following tests to be executed as listed below. This provides the end user a pass fail status for the array. The below test suite provides a comprehensive functional test of the arrays (data retention and other stress tests not included). The series of tests in gonogo are the culmination of internal ARM memory test engineers' past experiences in memory test.

[0120] 1) Perform wckbd 5's

[0121] 2) Rckbd 5's

[0122] 3) Wckbd a's

[0123] 4) Rckbd a's

[0124] 5) RWR_Ymarch 6's

[0125] 6) Bang 0's

[0126] At test instruction load, the user specifies the root dataword used during the test algorithm (exception during go/nogo BIST algorithm). This allows the user to select values stored into initialised arrays (iddq or ATPG) or to select new datawords to search for unexpected sensitivities during march and bitline stress tests.

[0127] The memory interface is defined in the below table. Its use is required for all memory devices on a core, regardless of whether or not ARM provides the memory for that core. This will allow the user to make use of the ARM BIST controller without special modifications. ATPG related support is also defined here.

Signal	I/O . Size	Description
MTestSel	In, <:0>	Optional, Used for serial test of
		multiple downstream arrays
MTestWEx	In	Dictates write operation
MTestREx	In	Dictates read operation
MTestADDR*	In, <maxAddr:0>	Target Address
MTestDINx	In, <datawidth:0>	BIST Write Data
MTestDOUTy	Out, <datawidth:0>	BIST read Data
MTestON	In	Global Override of cache logic
ArrayDisable	In	Blocks all array activity,
		regardless of operation mode
*May have single Address bus if mBIST_dispatch handles all scrambling concerns. This is still being thought out.

[0128] MTestSel<:0>

[0129] Should a memory BIST interface support more than one downstream array, separate enables are required to select array being targeted. This signal may also be required for a single array that has multiple repairable areas. A single array interface does not require use or porting of this signal.

[0130] MTestWEx

[0131] This signal (if not blocked internally by MTestSel or MTestON) indicates a write operation is being delivered to the memory test array interface. Address and Data information will be delivered coincident with this signal. If a write operation takes two cycles to complete, this signal is designated MTestWE2. If multiple arrays are supported by the interface, all with varying pipeline delays, the x designation shall be used (MTestWEx). Use of the pipeline delay length in the signal name allows for easier BIST integration by the BIST interface unit or end users own implementation.

[0132] MTestREx

[0133] Read enable for array under test. See MTestWEx signal for further information.

[0134] MTestADDR<:0>*

[0135] Address being targeted in array under test. Signal is as wide as maximum number of addressable elements in targeted arrays.

[0136] MTestDINx, MTestDOUTy

[0137] Data in and data out ports for BIST operations. MBISTDOUT delivered xy cycles later as defined by MBISTREx. Example: An MBISTRE3 specifies that the read operation will occur 3 clocks after MBISTRE3 information is asserted. The dataout pipeline may have its own offset and is represented by y. If multiple arrays are supported by a single interface, and some arrays have smaller data widths than the maximum array width, the memory BIST interface shall be required to deliver 0 on all unused data bits for all read operations. If datain and dataout busses have varying pipeline offsets depending on downstream array, they shall have an offset designated by x.

[0138] MTestON

[0139] This is a global override signal, informing the interface and downstream logic that BIST operations are imminent. This signal is the second highest priority signal in cache logic, behind ScanMode/ScanEn. Normal function/mission mode operation is a LOWER priority operation.

[0140] ArrayDisable

[0141] ATPG signal, blocking writes and reads to all arrays whenever active. Blocking must occur after all scannable state elements to insure safety of array's initialized data. This signal is generated in the memory BIST dispatch unit and is the or'd function of IDDQtest and ScanEn. IDDQtest used to insure preservation of state during iddq scan and capture, scanEn used to preserve state during ATPG load/unload should ATPG ram tests be implemented.

[0142] May have single Address bus if mBIST_dispatch handles all scrambling concerns. This is still being thought out.

[0143] MTestON overrides all normal function operations.

[0144] MTestON is asserted at least 4 cycles prior to first operation and not released until after last operation completes. Remains static on for duration of BIST test.

[0145] ArrayDisable blocks all write/read activity, regardless of device mode (functional, ATPG, BIST, etc.) and occurs after last scannable state elements.

[0146] A BIST no-op (no operation) can and will occur. Defined as BIST cycle where RE/WE are both inactive

[0147] Back to back operations must be supported and expected in any combination of R/W/Addr

[0148] All operation data provided from F/F's and read data will be sampled into F/F's. (mBIST is a flop based design)

[0149] All BIST operations can occur at full speed.

[0150] Addr and RE/WE signals are provided in same clock cycle. DataIn may be injected at different stage if required.

[0151] All RE/WE/Data signals must have cycle suffix designator as specified in signal descriptions.

[0152] All read operations must be completed within 8 cycles from RE assertion to dataout. (pipeline injection points must be reasonable).

[0153] If multiple arrays are supported by interface, address and data adjustments will occur in dispatch unit (example: addressed array is smaller than BIST_engine . . . BIST_dispatch will block RE/WE for out of range addresses)

[0154] All Variable sized arrays must be right justified in addressing wrt to incoming BISTADDR where rightmost bit is LSB.

[0155] ONLY ArrayDisable and MTestON can be assumed to toggle without regard to operation setup/hold times. All other control signals at interface must be considered unknown from cycle to cycle.

[0156] No address scrambling allowed between mBIST interface and actual array interface. Scrambling will occur in the dispatch unit.

[0157] Physical mapping characteristics must be maintained on all arrays downstream of each unique array interface. (Xaddr LSB always selects between adjacent rows regardless of array supported by gi9ven interface).

[0158] A dispatch unit is required between every BIST controller to memory test interface. This block serves as glue logic between a standardized front end BIST controller and a standardized memory test interface. It also provides manual debug accesses to the array and performs data compare and datalogging of the read data from the test interface.

[0159] The following table describes the incoming and outgoing interface signals of the mBIST_dispatch unit:

Input/O Data from Memory BIST controller	I/Output to memory test interface
Signal	Description	Signal	Description
ArrayEn<:0>	Enables target arrays (instr reg)	MTestSel	Array enables to interface
BISTTest	Game on	MTestWEx	Array write enable
MBISTData<3:0>	Write/read data	MTestREx	Array Read Enable
MBISTXaddr<:0>	BIST Controller Wordline Addr	MTestADDR*	Array cell select
MBISTYaddr<:0>	BIST Controller bitline Addr
MBISTWE	BIST Controller write enable	MTestDINx	Array Data in (full width)
MBISTRE	BIST Controller read enable	MTestDOUTy	Array Data out (full width)
MBISTShiftI	Shiftenable for dispatch instr	MTestON	Array test override
MBISTShiftD	Shiftenable for datalog	ArrayDisable	Array access blocking
DispatchDin	Shift data in
DispatchDout	Shift data out
IddqTest	IddqTest mode
SE	Scan enable input
SCANMODE	ATPG testing mode
MBISTFail<:0>	Fail bits sent to main controller
Notes:
1. BISTData<3:0> is linearally expanded across full memory bit width. Pipelining of this data inside of dispatch unit is done as <3:0> and expanded as needed. This expanded word is used to drive or compare the full bus width of the memory under test.
2. DispatchDout follows LSB of dispatch unit's instr or datalog register (as selected by shiftenable)
3. The memory test interface shall be driven by the dispatch unit and ONLY the dispatch unit for all defined signals (no direct connect to BIST controller as dispatch unit pipelines controller data for timing safety).
4. Support of instruction register capability inside dispatch is design specific. Datalogging/Repair information is generated and maintained in dispatch unit and is design implementation specific.
5. Multiple enable selected arrays can be tested simultaneously by mapping enables to higher order Yaddress. This is highly desirable for go/nogo testing.

[0160] Hard cores will have known physical to logical mapping of address spaces and will need to be scrambled accordingly in the BIST_dispatch unit. This is performed at the dispatch unit as address scrambling requirements may differ for different memory test interfaces.

[0161] Softcores and compiled rams will not have known addressing up front and scrambling cannot be performed in rtl, especially if the core design is synthesized prior to compilation of memories. Scrambling will be required to insure physical mapping exists with the ARM BIST algorithms. This is accomplished by adding 2 additional ports of the TestADDR to the dispatch unit. These ports represent the BIST controller address and the second represents the scrambled version of address going to the MtestAddr interface. These ports are brought up to the highest hardened level of the design and scrambling occurs by stitching the controller address to the test address as required. This stitching becomes the responsibility of the core integrator.

[0162] Although illustrative embodiments of the invention have been described in detail herein with reference to the accompanying drawings, it is to be understood that the invention is not limited to those precise embodiments, and that various changes and modifications can be effected therein by one skilled in the art without departing from the scope and spirit of the invention as defined by the appended claims.