15.01.2013 Views

U. Glaeser

U. Glaeser

U. Glaeser

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

FIGURE 45.2<br />

Input formats of MILEF are ISCAS-benchmarks (for circuits with exclusive primitive gates) and SPICE<br />

for circuits including complex gates, transmission gates, and bus-structures.<br />

Transistor-level information is actually reduced to a simplified switch-level structure for the test<br />

generation. All transistor-level networks are first analyzed by an advanced logic extraction program,<br />

which recognizes structures representing primitive gates such as NANDs, NORs, and inverters [31].<br />

Transistor-level networks that contain only simple gates are shifted to the gate-level before any explicit<br />

test generation and are therefore not dealt with at the switch-level. Local test generation is done for<br />

nontrivial macros only and can mostly be limited to one-stage complex gates and smaller networks.<br />

The example circuit in Fig. 45.2 consists of a number of primitive gates and one transistor netlist. This<br />

transistor netlist is divided into two parts by the extractor. The first part of this transistor netlist can be<br />

identified as a 2-input NAND and is therefore shifted to gate-level by the extractor. Test generation for<br />

this gate is done exclusively at gate-level. The second part of the transistor netlist representing a ANDNOR<br />

function could not be mapped to any primitive gate. Consequently, test generation for this gate is done<br />

at the switch-level. Fault propagation over this switch-level macro during the test generation is done at<br />

gate-level by using the 0- and 1-cubes of the switch-level macro. The idea of extracting gate-level modules<br />

from switch-level circuits was already proposed in [32].<br />

To also handle sequential circuits, the extraction algorithm can also identify several sequential elements<br />

such as, for instance, flip-flops and D-latches.<br />

Switch-Level Test Generation<br />

The local test patterns for the switch-level macros remaining after the extraction are generated by<br />

CTEST (CMOS test pattern generator) [30,33]. CTEST generates transition test pattern pairs with a<br />

hamming distance 1 between initialization and test. Therefore, the robust stuck-open test condition is<br />

satisfied for local test patterns since the switch-level macros extracted do not have internal path reconvergencies.<br />

To satisfy this condition also globally in as many cases as possible, MILEF avoids static hazards<br />

at the inputs of the switch-level macro with the actual fault whenever possible.<br />

CTEST has a reasonable performance for circuits up to the size of 200 transistors. This is sufficient<br />

for the MILEF approach, because the switch-level macros left by the extractor are relatively small, in<br />

general, not exceeding 20 transistors [31]. The time spent for switch-level test generation measured in<br />

experimental results was less than 5% of the overall test generation time, even when the extraction process<br />

© 2002 by CRC Press LLC<br />

I1 I2 I3 I4 I5 I6 2-NAND<br />

Logic extraction of switch-level macros.<br />

V DD<br />

OUT<br />

GND<br />

switch-level macro<br />

O 1<br />

O 2

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!