February_EDFA_Digital

edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 20 NO. 1 14 With the “passing” shift test sequence, the delay gen- erated by the failing flip-flop is "caught up" by the next flip-flop, due to theDATA-to-CK setup time, and is invisible. With the capture test sequence, the output signal of the failing flip-flop is stuck and propagates along the scan chain. The same behavior is observed on a failingmemory with a self-timed test sequence using the internal clock and a non-self-timed sequence. ELECTRICAL FAILURE MECHANISM When the flip-flop fails, a test sequence with a short clock pulsewidth is involved. The transition at the output is not generated because the clock goes low before the critical PMOS1 has finished switching. Figure 4 shows that the falling edge of the clock is acting as a timing clamp. If the signal generated by the critical PMOS is affected by a monograin switch before this “timing limit,” there is no fail and it is not possible to detect themonograin even if it is present. If the signal is too slow and switches after this limit, the transition is clamped before the signal reaches a high level, and a fail is generated. Themonograin defect is sensitive to the supply voltage and is detected here by clock pulse width variations. PHYSICAL ANALYSIS In the case of the scan chain, PMOS1 is clearly Fig. 5 (a) Layout including the fail and a reference PMOS for TEM lamella observation along the width of the transistor. (b, c) Nomorphological difference is observed inTEMandSTEM. (d) Onemainorientation (one color only) of thepolysilicon grain is revealed on the failing transistor in dark-field imaging. Fig. 4 Signal rising edge of a failing transistor impacted by a monograin defect. Failure is strongly linked to the pulse width of the clock. MONOGRAIN DEFECT IN POLYSILICON GATES (continued from page 12) (a) (b) (c) (d)

RkJQdWJsaXNoZXIy MjA4MTAy