Page 24

EETE MAR 2015

propagate to end point in a single cycle or not. Sometimes, it may. Sometimes, it may not. In other words, multi-cycle paths are a source of ‘X’. Hence, we can say that having multi-cycle and false paths in the design is not DFT friendly. In testing transition faults for ATPG testing, the number of X’s generated may be so high that it may reduce the efficiency of patterns generated by a significant amount. For transition faults testing using LBIST methodolgy, even a single MCP can be fatal as stated above. So, in LBIST, it is mandatory to break all the multi-cycle/false paths as shown in figure 3 below. For ATPG testing too, we can opt to break these paths to reduce the number of X’s generated. Dynamic x-bounding: As stated above, to achieve minimum possible test pattern count, and hence, test cost, all ‘X’ sources, including multi-cycle and false paths have to be masked. The masking of multi-cycle and false paths for X-propagation in test modes is termed as dynamic X-bounding. This can be achieved by either overriding the ‘X’ source with a control point or with the help of a bypass multiplexer. An alternative path is provided in test mode that provides the coverage for the required path. Theoretically, there can be different approaches for dynamic x-bounding depending upon the requirement for coverage and the effort that can be afforded as discussed below: Different approaches to dynamic x-bounding: Let us say, we have a multi-cycle path of general type “from start-point through through-point to end-point” as shown in figure 4. Please note that all the paths through through-point may or may not be multi-cycle. Similarly, all the paths starting from start-point may not be multi-cycle and all the paths ending at end-point may not be multi-cycle. Figure 4 shows all the possible combinations that can be possible in such a scenario. A number of approaches can be tried, each with its own pros and cons in terms of area, timing, power and test coverage as mentioned below: 1) Masking the data-path: In the simplest of the scenarios, all the paths through through-point can be masked by putting a ‘0-control point’ or ‘1-control point’ blocking all the paths that are multi-cycle DEBUGGING &PROGRAMING TOLS as shown in the figure 5. In this case, all the paths through the through-point will get blocked depriving of coverage of all the paths (whether multi-cycle or not) through the through-point which may be significant. This kind of x-masking can be deployed in LBIST transition testing as transition coverage requirement is very less. Using this approach in transition ATPG testing may prove to be fatal as a lot of paths are masked proving very costly for overall test budget. 2) Through-point masking/Endpoint masking approach: In this approach, we can consider transition testing as a twopass process. In first pass, the through-point is masked and the paths to end-point not passing through through-point are captured at the end-point as shown in figure 6. In second pass, the through-point is made transparent. In this pass, to disable the multi-cycle paths, the scan architecture is made such that when through-point is made transparent, end-point goes in shift mode propagating the data at its scan-in pin to its output as shown in figure 7. Thus, in pass 1 as shown in figure 6, we have to time the end-point’s scan-in path at-speed which can be difficult to meet if functional frequency is very high. Also, there is an introduction of one extra logic level in scan-enable path, which can also hurt if functional frequency is very high. Following this approach, we lose the coverage of all the paths through through-point and going to end-point regardless of whether they are multi-cycle or not. 3) Through-point masking and end-point clock masking: In a variation of case 2, the clock of the end-point can be masked instead of keeping it in shift mode even during capture phase as shown in figure 8. This will save from timing scan paths at-speed, but, the output of the end-point will remain constant during capture depriving of its coverage. Thus, there are supposed to be handful extra patterns required than case 2. This approach will Fig. 1: X propagation from non-scan flop/latch. Fig. 2: The probable X generating latch output is x-bound. Fig. 3: X-bound on exceptions/multi-cycled path which are single cycle in DFT modes. Fig. 4: Typical multicycle path scenario in soCs (from a startpoint ➔ through through-point ➔ to endpoint as highlighted. 24 Electronic Engineering Times Europe March 2015 www.electronics-eetimes.com


EETE MAR 2015
To see the actual publication please follow the link above