A digital twin is a digital representation of a real-world item, and includes software objects or models that represents these real-world items. In MEMS product development, digital twins (or software models of a MEMS device) can be used to represent a physical MEMS device and can minimize physical prototyping by use of predictive software models. These models are not only valuable in the conceptual phase of design, but during all levels of product development such as MEMS device design, packaging, IC design and system design. Digital twins enable faster product development by supporting virtual product testing and experimentation, and by minimizing physical prototypes, sequential development and long build-and-test cycles.
The MEMS industry typically employs a stage-gate methodology [1] to develop new products. This methodology is composed of five stages, starting at the concept and feasibility stage and ending in volume production (see Figure 1). The methodology can look slightly different for integrated device manufacturers and fabless companies, but in all cases product development can move into the next stage only after passing a “gate” with well-defined milestones and success criteria.
Figure 1: Stage-Gate Methodology during MEMS Product Development
While this methodology is not specific to MEMS design, one characteristic of conventional MEMS product development is that the development cycles are long (historically, at least 5 years) Even after this duration, successful yield on silicon is not at all guaranteed.
In many MEMS development projects, these five stages continue to be dominated by a “build-and-test” approach, leading to long and unpredictable development cycles. The alternative to this approach is to employ digital twins (or predictive device models) to “virtualize” the product development process. These predictive models are valuable during all stages of the development process.
At the proof-of-concept phase, most MEMS engineers use a simplified analytical model to investigate the ideal behavior of their device. In a research environment, the engineer might also have a working MEMS device constructed in silicon, but it is usually far from being a manufacturable product. A digital twin, in the form of a process-sensitive model, enables at this stage to combine the advantages of an analytical model with early experimentation and testing. The digital prototype can verify the proof-of-concept device, and do so much faster than using prototypes produced in a fab.
At the design development stage, the initial version of the MEMS design is brought to the next level through the production of advanced prototypes. Process flow development, IC design and package design are often improved in a parallel process, to create the advanced MEMS prototype. This stage includes many design iterations, with each design iteration taking several months since each prototype must be built in a fab and then fully characterized. These design iterations can be accelerated by running virtual “Design of Experiments” (DOEs) within a predictive MEMS model, avoiding lengthy fabrication and test cycles. Virtual experiments also provide early and important insight into MEMS device manufacturability. Moreover, a predictive process-sensitive model allows investigations that are impossible during normal fabrication, including the testing of an unlimited number of design permutations and deep exploration of complex device behavior. The same type of model can also be used to investigate package effects on the MEMS device and perform initial MEMS & IC co-design.
During this stage of product development, engineers need to optimize MEMS design and process, and subsequently freeze their design and pass it to pilot production. This typically requires many short loops of wafer fabrication to co-optimize their design and the available manufacturing technology. Designing a device into an established manufacturing process can substantially shorten this stage of development. A MEMS device model, once calibrated with actual data from a fabricated device, allows engineers to study process corners and perform sensitivity analysis on their design. The development of electronic read-out circuitry typically proceeds in parallel during the technology engineering phase, using a validated MEMS device model.
Once the design and the process flow are finalized, the development enters pilot production. This stage of development requires the fabrication of a massive number of product wafers. These wafers are used to stabilize the process and determine acceptable process tolerances that can maximize device performance and yield. Digital twins are extremely valuable in this stage of development. Predictive process models can provide insight into any trade-offs between process parameter variability and final device performance and yield, and thus accelerate the transition to the final stage of production.
Regular manufacturing of the product starts at this stage. Quality monitoring has been implemented, and a selection of process parameters is frequently measured to track yield and to guarantee that the device meets the final product specifications. At this stage, any digital twin should be validated and calibrated, to provide a complete understanding of device behavior while fully supporting failure mode analysis.
CoventorMP® is a software platform that can be used in MEMS product development to create a digital twin of an actual MEMS device. It can produce predictive software models that are valuable at all stages of product development. These models can be used during initial feasibility studies, process-sensitive design optimization and system and package-level co-design and development.
Figure 2: Digital twins developed using CoventorMP® enable initial feasibility studies, process-sensitive design optimization and system-level simulation
Digital Twins with predictive device models have many advantages over the traditional “build and test” MEMS development approach:
[1] Fitzgerald A.M., White C.D., Chung C.C. (2021) Stages of MEMS Product Development. In: MEMS Product Development. Microsystems and Nanosystems. Springer, Cham. https://doi.org/10.1007/978-3-030-61709-7_3