A Deep Learning-Based Approach to Uncertainty Quantification for Polysilicon MEMS

The path towards miniaturization for micro-electro-mechanical systems (MEMS) has recently increased the effects of stochastic variability at the (sub)micron scale on the overall performance of the devices. We recently proposed and designed an on-chip testing device to characterize two sources of variability that majorly affect the scattering in response to the external actions of inertial (statically determinate) micromachines: the morphology of the polysilicon film constituting the movable parts of the device, and the environment-affected over-etch linked to the microfabrication process. A fully stochastic model of the entire device has been set to account for these two sources on the measurable response of the devices, e.g., in terms of the relevant C-V curves up to pull-in. A complexity in the mentioned model is represented by the need to assess the stochastic (local) stiffness of polysilicon, depending on its unknown (local) microstructure. In this work, we discuss a deep learning approach to the micromechanical characterization of polysilicon films, based on densely connected neural networks (NNs). Such NNs extract relevant features of the polysilicon morphology from SEM-like Voronoi tessellation-based digital microstructures. The NN-based model or surrogate is shown to correctly catch size effects at a varying ratio between the characteristic size of the structural components of the device, and the morphology-induced length scale of the aggregate of silicon grains. This property of the model looks to indeed be necessary to prove the generalization capability of the learning process, and to next feed Monte Carlo simulations resting on the model of the entire device.

Download Full-text

Stochastic Mechanical Characterization of Polysilicon MEMS: A Deep Learning Approach

Proceedings ◽

10.3390/ecsa-6-06574 ◽

2019 ◽

Vol 42 (1) ◽

pp. 8

Author(s):

José Pablo Quesada Molina ◽

Luca Rosafalco ◽

Stefano Mariani

Keyword(s):

Deep Learning ◽

Learning Strategies ◽

Mechanical Response ◽

Learning Strategy ◽

Monte Carlo Analysis ◽

Small Samples ◽

Micro Electro Mechanical Systems ◽

Polysilicon Film ◽

Polysilicon Films

Deep Learning strategies recently emerged as powerful tools for the characterization of heterogeneous materials. In this work, we discuss an approach for the characterization of the mechanical response of polysilicon films that typically constitute the movable structures of micro-electro-mechanical systems (MEMS). A dataset of microstructures is digitally generated and a neural network is trained to provide the appropriate scattering in the values of the overall stiffness (in terms of the Young’s modulus) of the grain aggregate. Since results are framed within a stochastic procedure, the aim of the learning strategy is not to accurately reproduce the microstructure-informed response of the polysilicon film, but instead to provide a fast tool to be used at the device level for Monte Carlo analysis of the relevant performance indices. Accuracy of the proposed approach is assessed for very small samples of the polycrystalline aggregate to check if size effects are correctly captured.

Download Full-text

A Stochastic Model to Describe the Scattering in the Response of Polysilicon MEMS

Engineering Proceedings ◽

10.3390/engproc2020002095 ◽

2021 ◽

Vol 2 (1) ◽

pp. 95

Author(s):

Luca Dassi ◽

Marco Merola ◽

Eleonora Riva ◽

Angelo Santalucia ◽

Andrea Venturelli ◽

...

Keyword(s):

Silicon Film ◽

Micro Fabrication ◽

Micro Electro Mechanical Systems ◽

Local Fluctuations ◽

Stochastic Framework ◽

On Line ◽

Chip Testing ◽

On Chip ◽

Time Required ◽

Polycrystalline Silicon Film

The current miniaturization trend in the market of inertial microsystems is leading to movable device parts with sizes comparable to the characteristic length-scale of the polycrystalline silicon film morphology. The relevant output of micro electro-mechanical systems (MEMS) is thus more and more affected by a scattering, induced by features resulting from the micro-fabrication process. We recently proposed an on-chip testing device, specifically designed to enhance the aforementioned scattering in compliance with fabrication constraints. We proved that the experimentally measured scattering cannot be described by allowing only for the morphology-affected mechanical properties of the silicon films, and etch defects must be properly accounted for too. In this work, we discuss a fully stochastic framework allowing for the local fluctuations of the stiffness and of the etch-affected geometry of the silicon film. The provided semi-analytical solution is shown to catch efficiently the measured scattering in the C-V plots collected through the test structure. This approach opens up the possibility to learn on-line specific features of the devices, and to reduce the time required for their calibration.

Download Full-text

Micromechanical Characterization of Polysilicon Films through On-Chip Tests

Sensors ◽

10.3390/s16081191 ◽

2016 ◽

Vol 16 (8) ◽

pp. 1191 ◽

Cited By ~ 12

Author(s):

Ramin Mirzazadeh ◽

Saeed Eftekhar Azam ◽

Stefano Mariani

Keyword(s):

Polysilicon Films ◽

On Chip ◽

Micromechanical Characterization

Download Full-text

Closed-loop feedback control of microfluidic cell manipulation via deep-learning integrated sensor networks

Lab on a Chip ◽

10.1039/d1lc00076d ◽

2021 ◽

Author(s):

Ningquan Wang ◽

Ruxiu Liu ◽

Norh Asmare ◽

Chia-Heng Chu ◽

Ozgun Civelekoglu ◽

...

Keyword(s):

Deep Learning ◽

Sensor Networks ◽

Feedback Control ◽

Closed Loop ◽

Microfluidic System ◽

Cell Manipulation ◽

Integrated Sensor ◽

Loop Feedback ◽

Closed Loop Feedback ◽

On Chip

An adaptive microfluidic system changing its operational state in real-time based on cell measurements through an on-chip electrical sensor network.

Download Full-text

Simba

Communications of the ACM ◽

10.1145/3460227 ◽

2021 ◽

Vol 64 (6) ◽

pp. 107-116

Author(s):

Yakun Sophia Shao ◽

Jason Cemons ◽

Rangharajan Venkatesan ◽

Brian Zimmer ◽

Matthew Fojtik ◽

...

Keyword(s):

Deep Learning ◽

Large Scale ◽

Data Locality ◽

Coarse Grained ◽

Batch Size ◽

Peak Performance ◽

Large Scale Systems ◽

High Area ◽

On Chip ◽

And Storage

Package-level integration using multi-chip-modules (MCMs) is a promising approach for building large-scale systems. Compared to a large monolithic die, an MCM combines many smaller chiplets into a larger system, substantially reducing fabrication and design costs. Current MCMs typically only contain a handful of coarse-grained large chiplets due to the high area, performance, and energy overheads associated with inter-chiplet communication. This work investigates and quantifies the costs and benefits of using MCMs with finegrained chiplets for deep learning inference, an application domain with large compute and on-chip storage requirements. To evaluate the approach, we architected, implemented, fabricated, and tested Simba, a 36-chiplet prototype MCM system for deep-learning inference. Each chiplet achieves 4 TOPS peak performance, and the 36-chiplet MCM package achieves up to 128 TOPS and up to 6.1 TOPS/W. The MCM is configurable to support a flexible mapping of DNN layers to the distributed compute and storage units. To mitigate inter-chiplet communication overheads, we introduce three tiling optimizations that improve data locality. These optimizations achieve up to 16% speedup compared to the baseline layer mapping. Our evaluation shows that Simba can process 1988 images/s running ResNet-50 with a batch size of one, delivering an inference latency of 0.50 ms.

Download Full-text

On-Device Deep Learning Inference for System-on-Chip (SoC) Architectures

Electronics ◽

10.3390/electronics10060689 ◽

2021 ◽

Vol 10 (6) ◽

pp. 689

Author(s):

Tom Springer ◽

Elia Eiroa-Lledo ◽

Elizabeth Stevens ◽

Erik Linstead

Keyword(s):

Machine Learning ◽

Deep Learning ◽

Real Time ◽

Operating Systems ◽

System On Chip ◽

Low Latency ◽

Management Framework ◽

On Chip ◽

Specialized Hardware ◽

Deterministic Behavior

As machine learning becomes ubiquitous, the need to deploy models on real-time, embedded systems will become increasingly critical. This is especially true for deep learning solutions, whose large models pose interesting challenges for target architectures at the “edge” that are resource-constrained. The realization of machine learning, and deep learning, is being driven by the availability of specialized hardware, such as system-on-chip solutions, which provide some alleviation of constraints. Equally important, however, are the operating systems that run on this hardware, and specifically the ability to leverage commercial real-time operating systems which, unlike general purpose operating systems such as Linux, can provide the low-latency, deterministic execution required for embedded, and potentially safety-critical, applications at the edge. Despite this, studies considering the integration of real-time operating systems, specialized hardware, and machine learning/deep learning algorithms remain limited. In particular, better mechanisms for real-time scheduling in the context of machine learning applications will prove to be critical as these technologies move to the edge. In order to address some of these challenges, we present a resource management framework designed to provide a dynamic on-device approach to the allocation and scheduling of limited resources in a real-time processing environment. These types of mechanisms are necessary to support the deterministic behavior required by the control components contained in the edge nodes. To validate the effectiveness of our approach, we applied rigorous schedulability analysis to a large set of randomly generated simulated task sets and then verified the most time critical applications, such as the control tasks which maintained low-latency deterministic behavior even during off-nominal conditions. The practicality of our scheduling framework was demonstrated by integrating it into a commercial real-time operating system (VxWorks) then running a typical deep learning image processing application to perform simple object detection. The results indicate that our proposed resource management framework can be leveraged to facilitate integration of machine learning algorithms with real-time operating systems and embedded platforms, including widely-used, industry-standard real-time operating systems.

Download Full-text