Technology is both becoming more interactive and connected, from Alexa to digital factories of the future. Sensors are the gateway to transforming the physical to the digital realm, yet they are often the limiting factor in the overall performance of the system. Sensor technology has exploded recently due to the advances in MEMS and semiconductor processes. These advances are helping to serve market demand at a very low cost. Ultrasonic fingerprint sensors are highly sophisticated and compact ultrasonic phased arrays, that have been adapted for manufacturer and semiconductor processes and compressed into a footprint that is small enough to fit into one cell phone. The most profound example of the explosion in sensor technology is the Internet of Things (IoT) movement. This is also known as “the trillion sensors initiative”. The movement predicts an ever-increasing amount of connected sensors in our environment. All this translates into a huge global market for sensors that is growing at a healthy rate year over year.
In this post we’re going to explore how simulation can accelerate IoT innovation and reduce design cycles by describing several real-world examples, which we ran in OnScale.
Piezoelectric (PZT) sensors are used in a wide range of applications from biomedical ultrasound to industrial measurement. Key design metrics for piezoelectric sensors typically relate to the sensor sensitivity and frequency of operation, along with its bandwidth and spatial beam width. Challenges arise from the large electromechanical domains to coupling between sensor structures and the load median which is typically a fluid or gas.
Above we have an example of a 3D simulation of a piezoelectric sensor operating in pulse echo mode. The sensor admits a pulse into the water load, which is reflected from a steel plate (shown here in red). The pulse then travels back to the sensor and is received as a voltage signal. With quarter symmetry applied, the model uses 20 million degrees of freedom and runs in around 25 minutes. It also includes “transmit” and “receive” electrical circuits, in order to provide an accurate picture matching the system electronics. Despite these features, the OnScale simulation only requires 2.3 gigabytes of ram. These results highlight the efficiency of OnScales solvers.
Register to watch our webinar to see the full demo of how this simulation can be set up and run in OnScale!
In the above example we have a simple PZT disk with its backing and matching layers, housed in a load medium (water in this case). We loaded this into OnScale’s Cloud scheduler and ran a design sweep. We varied the PZT thickness and the matching layer thickness in 20 steps each. This produced a total of 400 models, which all ran simultaneously in the cloud. We were able to run 400 3D simulations in only 25 minutes! Something that can take weeks in legacy packages.
But how do we use this? Our example varied PZT thickness from 8 to 12 millimeters and matching layer thickness from 2.6 to 3.2 millimeters. By displaying key design metrics for each of these designs, we can provide users with a very clear picture of the problem space allowing them to make better design decisions.
We have exported the study outputs from Matlab as an example of how they can be visualized. In this example, we have plotted the received voltage pulse and it’s spectrum, along with sensitivity, frequency and bandwidth for each design. By moving the cursor across the graph, we’re able to explore the space and gain important design insights. We found that it is clear that the PZT thickness directly affects the center frequency, while fine tuning the matching layer optimizes the bandwidth of the device.
PMUT – Fingerprint Imaging Array
For our next example we are going to take a look at how simulations can be used to assess the performance of Ultrasonic Fingerprint Imaging Sensors. These Devices use PMUTS as their primary transduction method, which consist of a diaphragm on a silicon substrate coated with piezoelectric material. PMUTs are used in a range of broadband sensing applications including gesture recognition. We’ll begin with a simple Piezoelectric example before looking at how PMUT simulations are supporting the development of imaging algorithms.
PMUT design (top) and 3D unit cell model geometry (bottom)
Register to watch a full webinar on this topic.
The model we are using in this example is based on a paper by David Harshly, which was presented at the 2016 IEEE ultrasonics symposium. The design consists of an aluminum nitride active layer under a silicon membrane. The design frequency is 14 megahertz and the full array is contained over 6,000 PMUT cells. We ran a simple 3D unit cell model that took only 80 seconds to run, yet provided us with a wide range of information on the fundamental device performance.
While single cell results are very useful, an ideal PMUT design work flow goes much further. OnScale’s ability to solve large problems allows designers to move beyond these initial models and begin to consider the effect of array configuration on imaging performance. Designers use OnScale to simulate PMUT arrays within the full display stack up to assess the impact of manufacturing tolerances on array imaging performance. At each stage, design sweeps allow designers to quickly arrive at the optimum design, from simple cell construction all the way through to coupling into the display stack.
Fingerprint phantom on glass substrate (left) 400 element PMUT array (right)
The output of PMUT array simulations can be fed directly to imaging algorithms to help improve their performance. In this example, we took an array with 400 sensors and a 20 megahertz center frequency coupled to glass substrate. We then applied a tissue fingerprint to the reverse side of the glass to act as a target. We ran 400 parallel simulations on the cloud, each with a different sensor acting as a transmitter. Each time the received signal on every element was stored building a large data set (sometimes referred to as a full Matrix capture). The images were then constructed from that data using Matlab to implement a simplified imaging algorithm. Since all of the imaging was done post simulation, array elements can easily be activated and deactivated to assess performance with different array configurations.
Sparse array image 20 Random Tx elements Dense array image 400 Tx elements (full array)
Here we can see two images of tissue phantoms. On the left image, only the data sets from a random distribution of 20 elements were used, effectively creating a sparse array. On the right, all 400 transmit data sets were used. Producing an image of significantly higher quality using simulation to assess the final performance of sensor systems offers the potential to significantly reduce both the cost of prototyping and the time required to achieve design wins.
If you enjoyed reading about how simulation with OnScale can help drive innovation in IoT space with some of the above examples, register to watch our full webinar to see more simulation examples of:
- CMUT – Medical Imaging
– Optimizing Array Design
- SAW Filters
– Rapid, full 3D SAW Simulation
- BAW Filters
– 3D Shape Optimization