The ability to quickly and reliably fabricate nanoscale pore arrays in ultrathin membranes such as silicon nitride (Si x N) is extremely important for the growing field of nanopore biosensing. Laser-based etching of thin Si x N membranes immersed in aqueous solutions has recently been demonstrated as a method to produce stable functional pores. Herein, the principal mechanism governing material etching and pore formation using light is investigated. It is found that the process is extremely sensitive to the relative content of Si over N atoms in the amorphous membrane, produced by chemical vapor deposition. Commonly, Si x N membranes are made to be Si-rich to increase their mechanical stability, which substantially reduces the material's bandgap and increases the density of Si-dangling bonds. Hence, even minimal batchto-batch variation may lead to remarkably different etch rates. It is shown that higher Si content results in orders of magnitude faster etching rates. This rate is further accelerated in an alkaline environment allowing on-demand controlled nanopore formation in about 10 s time even at low laser radiation intensities. These results highlight that photoactivation of the Si x N by the incident beam is critical to the chemical etching process and can be used to readily produce nanopore arrays at any specific location.

Nanopores are single-molecule sensors capable of detecting and quantifying a broad range of unlabeled biomolecules including DNA and proteins. Nanopores’ generic sensing principle has permitted the development of a vast range of biomolecular applications in genomics and proteomics, including single-molecule DNA sequencing and protein fingerprinting. Owing to their superior mechanical and electrical stability, many of the recent studies involved synthetic nanopores fabricated in thin solid-state membranes such as freestanding silicon nitride. However, to date, one of the bottlenecks in this field is the availability of a fast, reliable, and deterministic fabrication method capable of repeatedly forming small nanopores (i.e., sub 5 nm) in situ . Recently, it was demonstrated that a tightly focused laser beam can induce controlled etching of silicon nitride membranes suspended in buffered aqueous solutions. Herein, we demonstrate that nanopore laser drilling (LD) can produce nanopores deterministically to a prespecified size without user intervention. By optimizing the optical apparatus, and by designing a multistep control algorithm for the LD process, we demonstrate a fully automatic fabrication method for any user-defined nanopore size within minutes. The LD process results in a double bowl-shaped structure having a typical size of the laser point-spread function (PSF) at its openings. Numerical simulations of the characteristic LD nanopore shape provide conductance curves that fit the experimental result and support the idea that the pore is produced at the thinnest area formed by the back-to-back facings bowls. The presented LD fabrication method significantly enhances nanopore fabrication throughput and accuracy and hence can be adopted for a large range of biomolecular sensing applications.

No abstract

No abstract

Objective: Machine learning techniques have been used extensively for 12-lead electrocardiogram (ECG) analysis. For physiological time series, deep learning (DL) superiority to feature engineering (FE) approaches based on domain knowledge is still an open question. Moreover, it remains unclear whether combining DL with FE may improve performance. Methods: We considered three tasks intending to address these research gaps: cardiac arrhythmia diagnosis (multiclass-multilabel classification), atrial fibrillation risk prediction (binary classification), and age estimation (regression). We used an overall dataset of 2.3M 12-lead ECG recordings to train the following models for each task: i) a random forest taking the FE as input was trained as a classical machine learning approach; ii) an end-to-end DL model; and iii) a merged model of FE+DL. Results: FE yielded comparable results to DL while necessitating significantly less data for the two classification tasks and it was outperformed by DL for the regression task. For all tasks, merging FE with DL did not improve performance over DL alone. Conclusion: We found that for traditional 12-lead ECG based diagnosis tasks DL did not yield a meaningful improvement over FE, while it improved significantly the nontraditional regression task. We also found that combining FE with DL did not improve over DL alone which suggests that the FE were redundant with the features learned by DL. Significance: Our findings provides important recommendations on what machine learning strategy and data regime to chose with respect to the task at hand for the development of new machine learning models based on the 12-lead ECG.