May_EDFA_Digital
edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 20 NO. 2 22 boundary of the soft block failure. Next, CPA analysis was performed with the device under a repeating test at the failing condition close to the voltage of pass/fail boundary. Figure 12 shows where a CPA spot was detected in a bit- line decoder area with a positive CPA experiment. Later, electrical nanoprobing and physical analysis showed Si dislocations at the interface of channel SiGe and Si, causing source-drain leakage in PFET P6 (Fig. 13). Why did a leaky P6 cause a positive CPA spot near N6? N6 and P6 formed a voltage divider, as shown in Fig. 14. When the laser illuminated N6, it caused the N6 I on to be higher than normal, which compensated for the leakage in P6, allowing the circuit topass from its initial failing condition. CASE #4: LVI/LVP ANALYSIS OF AN SRAM SCAN CHAIN FAILURE Today’s SRAM devices usually include a simple scan chain, which is used for setting test conditions or inter- nal functionality. At the beginning stage of technology development, scan chain failures are often themain yield detractors and prevent further testing of SRAM functional- ity. The fourth case study involves a 32-bit configuration register scan chain failure in an SRAM device. Because a functional test could not identify the failing register, LVI/ LVP was used to isolate the exact fault location. First, LVI diagnosis was performed to map the clock and data signals for a reference scan chain, as shownon the left side of Fig. 15. The yellow clock signal and blue data signal are clearly distributed across the entire scan chain. Next, LVI Fig. 12 CPA analysis showing a CPA spot in a bit line decoder area. Fig. 13 Layout view with the leaky PFET and CPA spot labeled. Fig.14 The bit-line decoder schematic with leaky PFET and CPA spot (NFET) labeled. diagnosiswas performedon the failing scan chain, and the acquired LVI image is shown on the right side of Fig. 15. It is clear that the clock signal is fine, as it continues through the entire scan chain, while the data signal propagation stops at register 10. The high-magnification LVI image shown on the left side of Fig. 16 indicates that the data signal is still strong at point 4, but faint at point 6, and then disappears beyond that. LVPwaveforms were collected from points 4, 6, and 8, as shown on the right side of Fig. 16. The data signal is still strong at point 4, but is corrupted at point 6, consistent with the observation on the LVI image. The LVPwaveformcollected frompoint 8 is the clock signal, which is good. So, the fault location of the scan chain failure is isolated to a small area in register 10, suitable for PFA. Subsequent PFA found hollow contacts near point 6. CONCLUSION Due to the SRAM array architecture in which bitcells are accessed through a hierarchy of bit-line andword-line decoder schemes, andwith the introduction of scan chain design, SRAMdevices often suffer from logic type failures. For such failures, a functional test alone cannot localize the defective area. More detailed fault isolation work is required prior to destructive physical failure analysis. Many diagnostic techniques are available for the fault iso- lationwork. These techniques employ alter- natives for stimulating the SRAMdevice and detecting a response and therefore provide a means for localizing various faults for dif- ferent failure types. Thus, careful selectionof diagnostic techniques for a particular SRAM logic failure is important for an efficient and successful analysis. For certain cases, a com- bination of two or more selected diagnostic techniques may be beneficial.
Made with FlippingBook
RkJQdWJsaXNoZXIy MjA4MTAy