Scan chain element and associated method

[0001] 1. Technical Field of the Invention

[0002] The present invention relates to scan chains used in testing of integrated circuits, and more particularly to flip-flops for use in such scan chains.

[0003] The invention has been developed primarily for use with scan chains for testing system on chip (SoC) circuits, and will be described hereinafter with reference to this application. However, it will be appreciated that the invention can be applied to many other forms of integrated circuitry.

[0004] 2. Description of Related Art

[0005] During testing of system on chip (SoC) designs, a predetermined stream of data is clocked into a serially connected set of D-type flip-flops on the chip. Once all of the data is latched by the relevant flip-flops and therefore available to the relevant inputs throughout the circuit, a suitable clock signal is applied to the circuitry. The results are then clocked out of the scan chain for interpretation.

[0006] A prior art multiplexed D-type flip-flop is shown in FIG. 1. It includes a positive-edge triggered flip-flop 10 with an output Q and optional outputs QN (inverted Q) and SO (dedicated scan output). The input to the flip-flop 10 is selected via a multiplexer (MUX) 11, which accepts as inputs D (normal functional data) and TI (scan test input). The mode of the MUX 11 is controlled by global scan enable test signal TE. If TE is low on the rising edge of the clock, then the value on the D input is captured into the flip-flop and made available at Q after a small delay. Conversely, when TE is high on a rising clock edge, the value at TI is clocked into the flip-flop and, again, made available at Q after a small delay.

[0007] One difficulty with this arrangement is that SoC designs often include multiple, asynchronous clock domains. In the absence of effort to balance all clock trees in scan mode, an attempt to capture all outputs simultaneously can result in incorrect data being captured due to skew between clocks in different domains. This problem is known as “shoot-through.”

[0008] One way of reducing this problem is to restrict test pattern generation tools to only allow one scan clock to capture in a given test cycle. However, this requires an increase in the number of test patterns that need to be generated by the tool, which in turn has a detrimental impact on test vector usage and overall test time.

[0009] Embodiments of the present invention overcome the above-described shortcomings of prior scan testing techniques and provide a scan-chain element and corresponding scan test technique that are substantially impervious to shoot-through.

[0010] According to one aspect of the invention there is provided a scan chain element in an integrated circuit, the scan chain element including: a first latch connected to accept test data as an input; a second latch connected to accept the output of the first latch as an input; control logic for accepting a clock signal and a hold signal; the scan chain element being operable in a first mode such that the control logic is configured to supply the clock signal to the first latch and subsequently, in response to the hold signal, to supply the clock signal to the second latch to latch the data from the output of the first latch.

[0011] Another aspect of the invention provides a method of capturing data in a scan chain for testing circuitry, the scan chain including a plurality of scan chain elements, each of the scan chain elements including: a first latch connected to accept test data as an input; a second latch connected to accept the output of the first latch as an input; and control logic for accepting a clock signal and a hold signal; the method including the steps of: supplying the clock signal to the first latch, thereby to latch in the test data; asserting the hold signal; in response to the hold signal being asserted, supplying the clock signal to the second latch to latch the data from the output of the first latch.

[0021] Reference is now made to the Drawings wherein like reference characters denote like or similar parts throughout the various Figures. Referring now to FIG. 2, FIG. 2 shows how the flip-flop of FIG. 1 can be considered in terms of a master latch 20 and a slave latch 21. The MUX 11 is as described in relation to FIG. 1. The master latch 20 is transparent (i.e., follows the input continuously) when the clock signal is low, whilst the slave latch is transparent when the clock signal is high. Within the present specification, the word “open” is used synonymously with “transparent”, whilst “closed” is used synonymously with “latched.” For the purposes of the description, the link between the output of the master latch and the input of the slave latch is labeled “M”.

[0022] Referring to the signal timing portion of FIG. 3, at time A, the clock is low, meaning that the master latch is open and the slave latch is closed. The slave stores the state of the flip-flop, and the value at Q represents the state held in the slave.

[0023] At time B, the clock goes high and the input data (either D or TI, depending upon the value of TE) is captured into the master latch as it closes. Simultaneously, the slave latch becomes open and the value in the master latch is passed to Q. Changes to the input at this stage will have no impact on the value of M or Q.

[0024] At time C, the clock goes low, which closes the slave latch such that the output of Q is latched at its present value. Simultaneously, the master latch opens. Effectively, the state of the master latch has been passed on and latched in the slave latch.

[0025] FIG. 5 includes a pair of serially connected scan chain elements to illustrate the prior art problem of shoot-through. The first scan chain element 50 is in a first clock domain whilst the second scan-chain element 51 is in a second clock domain. The output of the first element 50 is connected to the data input D2 of the second element 51 via a buffer 52. The multiplexer in each of the elements is controlled by a common test enable signal “tst_scanenable”.

[0026] There are two phases to scan-testing of the integrated circuit of which the elements 50 and 51 are a part. First, data is serially clocked into elements 50 and 51 via the S1 and S2 inputs during a scan shift cycle. It will be appreciated that there will typically be many more elements in a scan chain before and after the two illustrated. After the final scan shift cycle (shown in FIG. 5), all of the data in the elements is available to the correct respective inputs of the integrated circuits under test. In the example illustrated, both the first element 50 and the second element 51 hold a value of 1.

[0027] The test then moves into a capture phase, in which data is captured from the D1 and D2 inputs to the multiplexer. As can be seen in the timing diagram of FIG. 5, commencement of the capture phase is initiated by the tst_scanenable signal going low. It will be noted that as this happens, the value of D1 is 0 and the value of D2 is 1. The value on D1 is safely captured into the first element 50 at the first rising clock edge 53 in the capture phase. In the absence of shoot-through, the second element should capture the value that was at D2 (which is the value of the output of Q1) at the time the capture is intended to take place. However, if the result of capturing D1 into the first element 50 is made available at the output of Q1 before the clock edge reaches the second element 51, then the latched value of the first element 50 will be captured into the second element D2. In the present case, by the time the clock edge rises at point 54 in the clk2 domain, the output of the first latch has already settled to value 0 and been made available to the input of the second element 51. The original 1 value that was to have been captured at D2 from the output of the first element is therefore missed.

[0028] In the example of FIG. 5, there are two (internal) clock domains: clk1 and clk2. Assuming that the top level pins that control these internal clocks are pulsed on the tester with exactly the same timing, clk2 is delayed relative to clk1 due to different clock buffering. FIG. 4 is a signal timing diagram of clock signals clk1 and clk2 showing how those clock signals exist in comparison to how those clock signals are seen by each of the flip-flops.

[0029] Shoot-through is a problem that can affect scan-shift, scan-capture or both, depending upon the particular circuit implementation. Scan shifting can be made safe by the insertion of a latch in the scan chain, where the scan path crosses clock domains. This type of latch is commonly referred to as a “lock-up latch”. A more insidious problem is that of shoot-through in the capture cycle, as described above. This is because in most devices, there will be many functional paths that cross between clock domains. The classic condition for shoot-through is where clock skew is greater than functional path delay. It should be noted that the problem of shoot-through usually only applies to testing, since the circuitry is designed to take clock skew and functional delay into account by virtue of its design.

[0030] Turning to FIG. 6, there is shown a scan-chain element 60 for substantially eliminating the effect of shoot-through during scan testing. The scan element is described in similar terms to that of the element in FIG. 2, and like features are designated using like numerals. The additional hardware feature of the element 60 is an AND gate 61 that accepts as inputs the CK signal for the element and a novel notHold signal.

[0031] It will be apparent to those skilled in the art of digital design that when the signal notHold is held high, the element 60 functions in exactly the same manner as the prior art element described in relation to FIG. 2 (save for a slight delay in CK reaching the second latch due to the AND gate 61). Conversely, when notHold is held low, the clock signal CK is effectively disconnected from the second latch.

[0032] The operation of the element of FIG. 6 is shown in FIG. 7. During the functional cycle (i.e., when the integrated circuit is performing its intended use in an end product), notHold is held high and tst_scanenable signal TE is held low. The CK signal is supplied to both the first and second latches. Shoot-through should not be an issue in this mode, since the circuit is carefully designed and analyzed to ensure it is robust in terms of functional clock domains.

[0033] During the scan-shift procedure, both notHold and TE are held high. In the illustrated case, it is assumed that shoot-through is not an issue during scan-shifts. This is because it is typical to arrange that only those latches clocked by a common clock within a clock domain are connected to a particular scan chain. At time S (the rising edge of CK) during the last clock cycle in the scan-shift cycle, all the scan elements are loaded with test data. The functional combinatorial logic then evaluates, generating test data ready for capture in the capture cycle.

[0034] At the start of the scan-capture cycle and whilst CK is low, notHold is taken low. This has the effect of isolating the state of the second latch. The second latch holds the data that is of interest during the scan-capture cycle. At time A, the clock CK goes high, which captures the data at the input D into the first latch. Since notHold is held low, the second latch still retains the data it captured earlier. In the context of a larger scan-chain, where D would be connected to the output of an earlier scan chain element, the master (first) and slave (second) latches are effectively separated and holding the response and stimulus data respectively.

[0035] With a prior art scan-chain element, the value captured into the second latch at time A would pass through the open master, potentially changing the value of Q and causing the logic to start to re-evaluate. This re-evaluation is undesirable, since it leads to races between clocks and data. If clock paths are slower than data paths, the wrong data can be captured.

[0036] Returning to FIG. 7, once the response of the logic under test has been safely captured into the master latch, it is transferred to the slave latch by taking notHold high. Time B may be several tens of nanoseconds after time A. The exact timing of B is not particularly important, but it should be greater than the maximum delay between the clock source at the pads and the clock sink at the scan elements.

[0037] At time C, the clock CK falls to a low value. The first latch is then open and the slave closes. Again, the timing of C is not critical, but it will typically be several tens of nanoseconds after time B using current technology.

[0038] The response data can then be read out through the scan chain as would normally be done in the prior art.

[0039] FIG. 8 shows an arrangement similar to that in FIG. 5, except that the shoot-through safe flip-flops 80 and 81 are used. As in FIG. 7, the timing diagram shows that during the capture cycle, the notHold signal is held low whilst data D1 and D2 are clocked into the master latches 82 and 83 of first and second flip-flops 80 and 81 respectively. Once this has happened, notHold is brought high 84, which clocks the latched data into the respective slave latches 85 and 86 of each of the flip-flops 80 and 81. The data can then be clocked out of the scan-chain in the usual way.

[0040] It will be appreciated that the shoot-through safe waveform described for the scan-capture cycle can also be implemented during the scan-shift cycle.

[0041] Whilst it would be desirable to apply the invention to all flip-flops in a given device, it will be appreciated that the significant cost in extra area taken up by the additional logic and routing of the notHold test signal makes this an unlikely solution. A more practical approach is to replace flip-flops on the output boundary of an IP block with the shoot-through safe flip-flop of the present invention, which will give much of the advantage but significantly less usage of space. However, the precise number of and proportion of shoot-through safe flip-flops used will vary depending upon the particular application requirements of each design.

[0042] Although an exemplary embodiment of the method and apparatus of the present invention has been illustrated in the accompanying Drawings and described in the foregoing Detailed Description, it is understood that the invention is not limited to the embodiment disclosed.

[0043] For example, it is understood that the gate 61 may be utilized in a scan chain element in which the master latch is transparent when the clock signal is high and the slave latch is transparent when the clock signal is low (i.e., in a scan element performing as a negative edge-triggered flip flop). Further, it is understood that the particular circuitry for implementing gate 61 may vary. It is further understood that circuitry other than logic AND gate 61 may be utilized so long as the input notHold may allow slave latch 21 to receive the input clock CK when input notHold is in a first logic state, and prevent the input clock CK from being received at slave latch 21 when input notHold is in a second logic state different from the first logic state.

[0044] The invention is thus capable of numerous rearrangements, modifications, and substitutions without departing from the spirit of the invention as set forth and defined by the following claims.