Reliability On The Rise In IC Design

Reliability has been an important factor in the semiconductor industry for decades. A closer look reveals three main priorities:

1. In the area of technology development and optimization, the microscopic mechanisms that lead to degradation must be identified and understood before they can be fixed. Microanalytical methods are used here as well as TCAD simulations.
2. If it’s not possible to further improve the technology at acceptable costs, the remaining uncertainty must be quantified. Wafer level reliability (WLR) tests are typically used to characterize the ways high and prolonged stresses affect the behavior of individual transistors. The results are used to build WLR models and to forecast device lifetimes; the latter are documented as part of the technology qualification.
3. IC designers must be proficient in the technology they have at their disposal. Taking reliability into account means that the relevant information must be made available in the process design kit (PDK). This is the basis for using design for reliability (DfR) methods to ensure that a specific IC based on a given technology fulfills its service-life goals for its target application.

Degradation mechanisms and how they interact change from one technology generation to the next. Back when the main concerns were hot carrier injection (HCI) on NMOS transistors and negative-bias temperature instability (NBTI) on PMOS transistors, the introduction of high-k metal gate technologies, for instance, led to increased positive-bias temperature instability (PBTI) on NMOS transistors. The smaller the technology nodes, the more pronounced certain aspects such as non-conducting HCI or off-state time-dependent dielectric breakdown (TDDB) become. When it comes to wiring, electromigration is still the most important mechanism from a design point of view.

Designers don’t necessarily need an in-depth understanding of the microscopic mechanisms. They can use DfR methods such as aging simulation and electromigration analysis to test the long-term stability of their designs in a virtual environment. The models required are readily available in PDKs for advanced nodes but are occasionally provided for larger technology nodes as well. However, to interpret the results of such analyses and take suitable countermeasures, designers have to be familiar with the triggering stresses and underlying dependencies involved. While it’s possible to solve electromigration problems through broader wires and multiple vias (provided sufficient space is available), the various mechanisms that influence transistor aging mean that there is no blanket answer for that aspect of ensuring long-term stability.

Depending on how the aging models that map bias temperature instability and HCI are structured, they can provide designers with important information about the extent to which each transistor instance ages. This makes it easier to spot certain constellations that can then be classed as weaknesses. Examples include varying degrees of degradation in differential pairs of transistors and marked changes to output transistors in amplification stages. In addition to what the aging simulations can say about how the circuit’s performance changes over time, these interim results generally provide concrete information about which aspects of the design must be optimized.

The trouble is that not all aging models are structured the same way. Depending on the target application – analog, digital, or HF, for example – they map the degradation of different transistor characteristics and properties, which results in models of varying complexity. Moreover, as the number and detail level of interim results also vary widely, designers can end up taking vastly different bits of information from reliability simulations.

Modeling specialists are struggling with the complexity of how to model aging while also coping with a shortage of skilled workers. This situation will be further exacerbated if yet more modeling is required due to the diversification of semiconductor technologies for special applications and the increased use of compound semiconductors such as GaN and SiC. What’s more, chiplets and the potential rise in integrated optical interconnects pose additional challenges in the area of photonic reliability.

Despite the challenges that lie ahead, recent advances have shown that the semiconductor industry is well on course to achieve more reliability in everything from technology development to IC design. Time will tell whether the specific requirements of individual design projects will continue to demand tailored solutions, or if a uniform approach is preferred.