The testability of a modified Booth multiplier

Abstract

This paper describes the generation of the test view of the multiplier-accumulator which is based on a modified Booth algorithm. It will be shown that an algorithm can be found that generates very quickly the test patterns as a function of the parameters of the module. The fault model used is an extension of the traditional stuck-at model towards switch level faults occuring in pass transistor logic. In order to achieve full testability, only slight modifications on the hardware of the multiplier-accumulator were necessary. It was even possible to generate a set of test patterns which is independent of the size of the multiplieraccumulator.

C-Testable modified-Booth multipliers

“…A comparison of the proposed C-testable design with the designs given in [16,17] is provided in Table 5. The proposed design has the significant characteristic that it is the only which can be used under any design environment (full custom, standard cells, etc.)…”

“…The testability of multipliers based on the modifiedBooth algorithm, in the restrictive case when the structure of most or all of its cells was known, was examined in [16][17][18]. In [16] the test set for a multiplieraccumulator based on the modified-Booth algorithm was given requiring the addition of 2 extra primary inputs.…”

“…In [16] the test set for a multiplieraccumulator based on the modified-Booth algorithm was given requiring the addition of 2 extra primary inputs. The fault model that was used in [16] is an extension of the stuck-at fault model towards switch level faults in pass transistor logic. The cells were implemented using the Piramid silicon compiler [19].…”

“…For the specific DCVS implementations of the multiplier cells, the multiplier is C-testable with 31 test patterns, detecting all detectable single stuck-at, stuck-on and stuck-open faults, when 6 extra primary inputs, 2 extra primary outputs and some extra logic are added. In [16][17][18] the hardware and delay overheads were not explicitly calculated.…”

“…The proposed design is Ctestable with 61 test vectors when the former is used and with 73 test vectors when the latter is used. At this point we remark that the C-testable designs proposed in [16][17][18]21] consider only the use of ripple carry adders at the last stage of the multiplication.…”