|
Diagnosis-Driven Evaluation of Design-for-Manufacturability
Wing Chiu Tam |
Shawn Blanton |
Aggressive scaling of CMOS devices has pushed the integrated circuit (IC) manufacturing process to the extreme, to the point where the challenges encountered during fabrication can no longer be addressed by manufacturing tweaks alone. To have acceptable yield, designers must be aware of these challenges and make appropriate compensations in the design. In other words, the design for manufacturability (DFM) era is certainly upon us. To perform DFM, the designers must be aware of what to look out for, i.e., they need to know what systemic issues can adversely affect yield. Foundries provide both required and recommended DFM design rules to aid designers in this regard. DFM rules are generally developed during process characterization, thus making it likely that a given rule may not be applicable to a particular design. Therefore, in contrast to a required design rule, many recommended DFM design rules are often optional, leaving it up to the designer to determine which rules are important to enforce in a given design.
Unfortunately, because of the area, performance and time-to-market constraints on the product, more often than not, the recommended DFM rules are followed in an ad-hoc manner. Obviously, selection criterion followed in this way is not optimal. Our work (called RADAR, Rule Assessment of Defect-Affected Regions) addresses this issue using a diagnosis-driven approach which is illustrated in Fig. 1. A large number of failed ICs are diagnosed to identify the x-y-z failure area for each chip. Measures of rule adherence/violation are then collected for both passing and failing layout areas. Statistical analyses (e.g., hypothesis testing, causation confirmation, etc.) are performed to evaluate the effectiveness of the DFM rules in preventing yield loss. The results of applying RADAR to fail data from a 90nm Nvidia is shown in Fig 2. Specifically, Fig. 2 shows the relative importance of various DFM rules. Examination of the plot reveals that enclosure rules are more important than density rules, which suggests that via issues may be causing more yield loss than density issues in this particular design/process combination.
 |
 |
Fig 1. Flow diagram implemented in RADAR. |
Fig 2.DFM rule ranking and interpretation based on statistical analyses using RADAR. |
|
