A major shakeup in reliability testing
10/01/2005
If there is one grand truth in the semiconductor industry, it is this: Simple device scaling to achieve performance gains is dead. Gains at future technology nodes will no longer come from simply shrinking devices by ratcheting down wavelengths. As a result, fabs are introducing new materials at mind-numbing rates. Incremental performance improvements from these new material systems began to surpass those from litho technology at the 90nm node. However, newer materials are often electrically unstable and unpredictable, thus driving the need for a greater volume of reliability testing to be done at more points in the development cycle.
More expensive and riskier R&D must now occur far earlier in the lifecycle of a new technology node. Recent consolidations of R&D at industry consortia and smaller IDMs are evidence of the new reality. The plugged-up funnel of major new innovations underscores that new materials and devices are more challenging to develop than previously expected. These innovations include characterizing extremely low leakage and tunneling currents, Si and gate-dielectric interface defects and related degradation, negative-bias temperature instability degradation, recovery mechanisms in pMOSFETs, and charge-trapping in high-k gate dielectrics.
The industry followed a similar path nearly 20 years ago when Si/SiO2/polysilicon/Al materials were the new frontier in materials science. Understanding the basic materials and their integration into a full process required substantial effort. Unfortunately, once that was solved, the industry largely neglected materials engineering because the Si/SiO2/polysilicon/Al materials system is so well behaved. With two decades of this stable materials experience, the industry had little need for a large cadre of materials physicists. This lack of demand caused the infrastructure needed to train new materials scientists - required to overcome the next wave of emerging materials challenges - to atrophy.
Thus, there is a shortage of skilled scientists who can develop new material systems, expand reliability test methodologies, and interpret the huge volume of data this entails. Unfortunately, the industry doesn’t have the luxury of waiting for the redevelopment of a materials physicist infrastructure - Moore’s Law is calling us back to the future.
In addition to the materials science juggernaut, device voltage scaling driven by low-power mobile devices compounds the industry’s problems. The ability of even the Si/SiO2/polysilicon/Al materials system to deliver stable DC operating points, such as threshold voltage over extended product lifetimes, is becoming more difficult, as is the ability to cope with new electrically unstable materials systems.
The decades-long complacency, driven by the stability of Si/SiO2/polysilicon/Al materials systems, has led to an inefficient testing model. Some reliability testing is done in R&D and process development, and then little or none is performed during the process integration and volume wafer-production phases. It is then reinstated when the die is packaged.
The net results of the materials changes and voltage scaling are twofold. First, reliability testing needs to be increased in R&D, process development, and technology development. And second, because the results will be more difficult to analyze due to the physics of the new materials and a lack of qualified materials engineers, facilities will need to rely on test suppliers not just for measurement boxes, but for physics and materials expertise to interpret results.
To stay competitive, fabs will have to rely more on suppliers who have in-depth expertise, not just in instrumentation and measurements, but also in material behavior. Such suppliers can rapidly develop new test methodologies and equipment, help integrate testing throughout the entire development cycle - from R&D to volume production - and extract actionable information and knowledge from obtained data. This is the model required for fabs to quickly ramp new processes to acceptable yields and profitability.
In addition, reliability testing done mostly at the package level will need to move upstream to volume production and be performed on-wafer, both in-line and end-of-line. Testing speed requirements 100× faster than currently available methods are also pushing the envelope. Right now, it is unclear if this is an extension of existing electrical testers, or if new toolsets will be needed in the fab.
Either way, reliability testing is set for a major shakeup. Surviving and prospering depends on a company’s ability to change and keep up with the times.
 |
For more information, contact Steven Weinzierl, applications engineering manager, at Keithley Instruments, Inc., 28775 Aurora Rd., Cleveland, OH 44139; e-mail [email protected].