b^⅛∙∙‰tfX,∙∙∙⅛)r,
and
1 — Gι, l∙2 ∙ ■ ∙ In)t-
This notation allows following minimization for finding the variation 1.
min∣∣Al-b∣∣∣.
(4.9)
we call this method ^-minimization method.
Note that it may not be possible to find the variations of all gates by this
method. For example in Figure 4.2, if we want to find another sensitizable path
that includes g⅛, we should fix f = 1 (none-controlling value) causing e — 0 and
g = 1. Thus, the transition cannot propagate on the line g and path Po is the
only path that includes the gates <⅛, g⅛ and g^. As a result, there is at most two
equations (falling and rising) that includes variation of the gates gɜ, g½ and g6∙, it
is impossible to find the variation of the three gates separately. We refer to such
cases as ambiguous gates.
4.3 Delay estimation using compressive sensing
Section 4.2 presents a system of linear equations to estimate variations of the
gates. However, the optimization problem in Equation 4.9 does not consider the
spatial correlation of the delay variations. Incorporating the spatial correlation in
46