Gentest An Automatic TestGeneration System for Sequential Circuits

Gentest: An Automatic Test-Generation System for Sequential Circuits Mohamed Abougabal, Wael Hermas, Tarek Saad, Rami Abielmona ELG 5194 Wednesday November 22, 2000 Prof. Sunil Das

Presentor: Mohamed Abougabal n n Introduction Gentest Definition Converting Sequential to Combinational circuit Gentest Architecture

Introduction to Gentest n Definition: – Gentest is an automatic Test-Generation System for sequential Circuits. n Gentest: – Was introduced because of the rapid growth of VLSI chips, we certainly know that these chips contains faults. Therefore we need Gentest to help us achieve the high fault coverage. – It Overcomes the inefficiency of previous ATG methods for sequential circuits by combining a sequential test generator with a concurrent fault simulator.

Test Generation Techniques (1) n 1. Manually write test vectors (write a sequence of values provided to the circuit input) – Disadvantage: n n It’s tedious and time-consuming process 2. BIST : Built-In-Self test Circuit The circuit itself create test vectors. The circuit generates random patterns that can thoroughly test a combinational circuit. – Disadvantage: n n It requires more circuitry. The cost of chip is higher because of the area. It causes the chip to be slower Finding the exact fault coverage for test vector is difficult.

Test Generation Techniques (2) n 3. ATG: Automatic Test Generation. – Given the description of a circuit , ATG programs deliver a set of vectors. A complete ATG algorithm guarantee finding a test for a fault if one exists. – Disadvantage: n n It’s not practical for every circuit Since most of the sequential circuits are practical circuits we can use Scan Design Techniques to convert sequential circuits into combinational circuits. The difficulty of sequential test generation lies on the large number of decision steps.

Converting Sequential to Combinational Circuits (1) n n The First Sequential Test Generation Algorithm were just an extension of the combinational test algorithm. Those Algorithm are: – D- Algorithm – Podem (Path-Oriented Decision making. ) n We can model Sequential circuits by two blocks (Combinational and Memory)

Converting Sequential to Combinational Circuits (2) Figure 1. Sequential circuit model

Iterative Logic Array Method (1) n Now use this same concept and expanding on it we can model a Sequential circuits into Combinational in time Domain and this method is known as The Iterative Logic Array Method. n In Figure 2: Each copy corresponds to the circuit behavior at one time frame. Therefore D-Algorithm and Podem can be applied because these are combinational circuits.

Iterative Logic Array Method (2) Figure 2. Iterative logic array model

Iterative Logic Array Method (3) n Problems: – Inefficient in dealing with repeated faults sites. – Excessive memory usage because the circuit must be saved for each time frame in test sequence. n Solution: – Mallela and Wu have introduced STG 2 (sequential test generator using the Back algorithm) – Then Gentest was introduced. Gentest is combination of STG 2 and Csim ( Concurrent fault simulator).

Gentest Architecture (1) n Functionality of: – STG 2: Generates a test sequence for every target fault selected by Gentest – Csim : Determines which faults are detected by the STG 2 generated test sequences. – Gentest: Submits the vectors created for fault detection to a fault simulator. n Now the overall operation of Gentest. – 1. The user can provide a fault list; otherwise, Gentest creates one after collapsing all equivalent faults. – 2. User can set a time limit or let it be generated by an experimental formula in SGT 2. – 3. User can provide some initializing vectors that Csim will simulate to initialize the circuit

Gentest Architecture (2) – 4. Gentest selects an untried and undetermined fault as a target fault for SGT 2. – 5. SGT 2 tries to generate a test sequence for the target fault. – 6. So by looking at steps 1 to 5. Test sequence can be generated in one of the three cases: n n Test sequence can be generated to detect the target fault The target fault can prove untestable Time limit can be reached. In either of these cases Gentest returns to step 4.

Presentor: Wael Hermas n n n Iterative Logic Array Model Podem Algorithm Putzolu and Roth Test Method STG 2 Discussion Back Algorithm

The Test Generator n Problems of previous test generation method – Iterative Logic Array Model – Podem Algorithm – Putzolu and Roth Test Method

Iterative Logic Array Model (1) n n n The following figure (refer to figure 3) shows a sequential circuit with a single stuck-at fault. Each “*” is the fault site in each time frame. Using the iterative logic array model, this fault has a repeated fault effect at every time frame.

Iterative Logic Array Model (2) Figure 3. General model of a test sequence

Podem Algorithm n n n Assume that the fault, in the previous example, can only be tested by a test sequence with eight test vectors Lets us say that the Podem algorithm needs to know that the propagation of fault effects should start after time frame 5 (Podem algorithm will stop at frame 4) since all the sensitized paths are blocked Without knowing the test sequence, having this information would be impossible, therefore we can conclude that Podem algorithm can be very inefficient in generating test sequence for sequential circuits.

Putzolu and Roth Test Method (1) n n Uses the two-step approach of the D-algorithm to generate the test sequence. Their method has the same problem as Podem at time frames 1 -4. To solve this problem, the five-valued model of the Dalgorithm was extended to be the nine-valued model Table 1 compares the five-valued and nine-valued models

Putzolu and Roth Test Method (2) Table 1. Comparison of 5 -valued and 9 -valued models

The nine-valued model n The nine-valued model lets us use single-path sensitization in the D-algorithm. – The nine-valued model is very inefficient during justification – The nine-valued model has the following efficiency problems – To justify an N-input gate's output value, there are N*N choices – No simple testability guidance can rank these N*N choices.

Sequential Test Generator (STG 2) n n Because of the efficiency problems in the previous test generators, a complete test generator STG 2, without excessive memory requirements, is developed. Uses the Back Algorithm and the Split value model

The Back Algorithm (1) n The Back Algorithm proceeds as follows: – the algorithm selects the primary output with the sensitized value – this is accomplished through a new testability measurement called derivability – the fault is not testable if no primary output can have a sensitized value – the algorithm justifies the values required at the current time frame – the test sequence has been found if the state input requirements of the current time frame match the initial state values

The Back Algorithm (2) n Continued: – the back algorithm justifies time frame by time frame (all the requirements of one time frame can be handled simultaneously) – during test generation, the circuit status has to contain at most two time frames: the current and previous time frames. – Hence, we can conclude the Back Algorithm's memory requirements is minimal.

Presentor: Tarek Saad n n n Fault simulation Sequential circuit fault simulation Nine-valued model Split value model Testability guidances

Fault Simulation Good machine A logic model of the circuit-under test without any faults inserted Faulty machines A logic model of the circuit-under-test with one or more fault models inserted Fault specification Defining the set of modeled faults and performing fault collapsing Fault insertion Selecting a subset of the faults to be simulated and creating the data structures to indicate the presence of the faults

Sequential Circuit Fault Simulation Figure 4. Sequential circuit test simulation model

Nine-valued Model (1) n n n The good machine and bad machine are two independent sub-circuits with the same primary inputs. It combines the values of the two sub circuits to reduce memory needed to store circuit status. In this arrangement the two circuits must be justified simultaneously. Figure 5. Realization of a Sequential circuit Figure 6. The nine-valued model

Nine-valued Model (2) Inefficiencies: n Justifying two independent sub circuits simultaneously increases the number of backtracks and the time wasted. n Split value algorithm will solve this problem.

Split Value Model (1) n n n Extension to nine-value model to improve it’s performance during justification. It splits the nine values into two sets of three values for the good, and bad machine. Separate storage for the circuit statuses, good/bad machine can be justified separately. Figure 7. The split value model

Split Value Model (2) n n Each machine is justified separately, we can use testability measurement of each circuit to rank available choices. Split model uses default value to reduce backtracks significantly. Definitions: Good value. Value at the good machine. Bad value. Value at the bad machine. Relation between good and bad value. Good/bad values can be 0, 1, x(don’t care), or z(high impedance). Sensitized lines. Lines with unequal good and bad values. Non-sensitized lines. Lines with equal good and bad value, not affected by fault site. Relation. Could be sensitized, non-sensitized, or unknown.

Split Value Model (3) Dynamic identification of relation information proceeds as follows: – – If all input of an element are non sensitized lines, then all outputs must be non-sensitized. If any output of an element is a sensitized line, then at least one input must be a sensitized line. Advantages: – The high impedance value is used in bus element containing circuits to avoid bus conflicts.

Testability Guidance Two testability measurements guide decision making in conventional test generators: Controllability: measures the difficulty of placing a specific logic (0 or 1) value on that line. Observability: measures. The difficulty of observing the fault effect at that line from the primary outputs. • Both are calculated based on the good machine. • Observability guides the sensitized-path selection while controllability guides the line value justification. • Back algorithm (as mentioned) uses drivrablitiy similar to observabilityand measures the dificulty of driving the fault effect from the fault site to a line.

Presentor: Rami Abielmona n n n Experimental results STG 2 Analysis Gentest Analysis Conclusions Future Work and Biographies

Experimental Results Sequential Test Generators n n STG 2 is an extension of STG 1, described by Mallela and Wu STG 2 uses the Back algorithm and the Split value model, while STG 1 utilizes the extended backtrace method and the nine-valued model. STG 1. 5 uses the D-algorithm and the Split value model. From the experimental results that will be presented next, STG 2 was chosen for implementation in Gentest, because it has shown the best fault coverage and total test-generation time.

Experimental Results STG 2 Analysis (1) n n The three sequential test generators were compared by live-testing them on a Convex C -1 computer using a myriad of circuits, the latters’ characteristics shown in Table 2 Looking at that table, a few definitions are in order: – – – Gates p 1, p 0 I/O pins FF Faults >>> >>> >>> The number of gates in the circuit; The primary inputs and outputs; Either input or output, but not both; The number of flip-flops in the circuit The number of faults in the circuit.

Experimental Results STG 2 Analysis (2) Table 2. Characteristics of the circuits

Experimental Results STG 2 Analysis (3) n n Table 3 shows the detailed results of the analysis of STG 2 The definitions are as follows: – Time n The sum of the test-generation time in seconds for all the target faults; – Detected/Untestable/Dropped n The number of tests in each category; – Efficiency n The sum of detected and untestable faults divided by the total faults.

Experimental Results STG 2 Analysis (4) Table 3. Results for STG 2

Experimental Results STG 2 Analysis (5) n n n Table 4 compares efficiency and time for all three test generators. As we can see from the results, STG 2 spent more time than STG 1. 5 for circuits c 499 and c 880, but overall, as the number of gates and faults in the circuit increase, the efficiency of STG 2 is much higher than that of STG 1. 5, with a lower time factor as well. The latter observation leads us to the conclusion that test-generation overall time is much better using STG 2, thus was chosen for incorporation in Gentest.

Experimental Results STG 2 Analysis (6) Table 4. Comparison for STG 1, STG 1. 5 and STG 2

Experimental Results Gentest Analysis (1) n n Another set of experiments were carried out on a Sun 3/60 workstation, using the Gentest generation algorithm The set of circuits implemented this time vary from the previous ones, but still are represented by the same characteristics as the previous ones (See table 5, for a brief reference).

Experimental Results Gentest Analysis (2) Table 5. Characteristics of the circuits

Experimental Results Gentest Analysis (3) n n Table 6 shows the detailed results of the analysis of Gentest The definitions are as follows: – Time n The sum of the test-generation time, fault-simulation time and communication time between the CPU and the circuit in seconds for all the target faults; – Trial/Untestable/Sequences n The number of tests or sequences in each category; – Vectors/Detected n The number of vectors is the size of the final test sequence, while the detected column lists the faults detected by CSim; – Fault Coverage n The sum of untestable faults identified by STG 2 and detected faults identified by Csim, divided by the total number of faults.

Experimental Results Gentest Analysis (4) Table 6. Results for Gentest

Experimental Results Gentest Analysis (5) n Observations of the results are as follows: – The number of sequences generated by STG 2 is always much smaller than the faults detected by CSim; – The number of runs needed by STG 2 to identify all the untestable faults cannot be reduced, hence is on the critical path of the total test-generation time; – The CPU time limit was set to two seconds per fault, thus for circuits “logic 6” and “mickey”, the fault coverage was very low due to the amount of target faults in these circuits; and – The total average fault coverage did increase as the Gentest algorithm was applied, as opposed to the results in table 3, which showed the circuits after STG 2 alone was applied.

Paper Conclusions n n n Gentest has proven to be a very efficient automatic test-generation system for sequential circuits It minimizes the total time needed for testgeneration, as well as maximizes the efficiency Gentest also demonstrates an improved, but not ideal, test fault coverage

Future Work n n n Future work on test-generation algorithms will concentrate on large functional units, and more complex circuitries To do so, the test-generation process has to be sped up, through the use of high-level models or functional units, such as instantiating behavioral models of a circuit (e. g. ALU, shift register, binary counter), rather than their gate- or even switch-level implementations Such an algorithm had not yet been well defined

Author Biographies