A generalizable codesigned platform for solid-state nanopore sensing beyond the capacitive-noise constraints

Abstract

Solid-state nanopores offer label-free, real-time single-molecule sensing, yet resolving rapidly translocating biomolecules requires high-bandwidth data acquisition, where increased high-frequency noise fundamentally limits reliable recovery of informative events. Here we establish a hardware-software co-designed nanopore sensing platform that overcomes this bandwidth-noise limitation by integrating scalable low-noise device engineering with deep learning-based signal reconstruction. A wafer-scale dielectric-engineering strategy using low-dielectric SU8 photoresist reduces total device capacitance to the picofarad regime and suppresses high-frequency noise by up to fivefold while maintaining facile, controllable and reproducible fabrication. This extends usable acquisition rate to 40 MHz and enables capture of fast molecular features. Combined with a reconstruction neural network trained on synthetic translocation signals embedded in experimentally measured noise, the platform recovers transient blockade sublevels while preserving temporal fidelity. Using engineered DNA molecules carrying dumbbell-like barcodes, we resolve nanometer-scale structural features on sub-microsecond timescales, and experimentally measure translocation velocity within the sub-5 nm regime. Dual-channel measurement on a single nanopore device further demonstrates transferability of the platform by showing robust cross-channel signal reconstruction across distinct baseline noise levels. This platform provides a general route for reliable recovery of previously inaccessible molecular information from high-bandwidth nanopore measurements.