← Research

MESS 2026 · Poster

Web-based Interactive Tool for Real-Time Analysis of Multi-day Nanoscale CMOS Defect Spectroscopy

Handling terabyte-scale measurement data

Yurii Chubenko, Semih Ramazanoglu, Alicja Michałowska-Forsyth · Institute of Electronics, TU Graz

Microelectronic Systems Symposium (OVE) · Vienna

One stress campaign writes more than a terabyte, and at some point the hard part stops being how to record it and becomes how to actually look at it. I built a single viewer that spans the whole experiment: zoomed out it shows five days at a glance, zoomed in it resolves one ~25 nA charge-trapping step. It changes scale by itself, so a terabyte stays open to free exploration with no memory limit.

When the data outgrows the tools

One total-ionizing-dose campaign on a nanoscale transistor runs for days and writes more than a terabyte: over ten thousand 400 kHz oscilloscope captures, plus continuous bias, temperature, dose and phase logs. The instruments could record all of it; the tools I had could not show it.

  • • Scope software opens one capture at a time.
  • • MATLAB and Jupyter want the whole dataset in memory before they draw anything.
  • • Nothing put the instruments on one timeline, so the streams had to be matched up by hand.

From acquisition to an interactive window

  1. 1

    Acquire

    Five instrument streams on one shared UTC clock.

  2. 2

    Store

    ≈1 TB of raw captures + every instrument log, indexed in one capture catalogue.

  3. 3

    Cache

    Offline multi-resolution pyramid (1k/10k/200k pts, Min-Max LTTB) — ≈200 GB on SSD.

  4. 4

    Serve

    A lightweight server streams only the resolution you need — many analysts at once, in the lab or worldwide over a secure tunnel.

  5. 5

    View

    Pan, zoom and filter in the browser; the view auto-switches resolution as you zoom (semantic zoom), responding in < 200 ms.

Five streams · one shared time-base

Waveforms

2× R&S RTO2044 (HiSLIP)

Source current (TIA/CH4), gate CH2 & drain CH3

Bias setpoints

Keysight B2962A SMU (LXI)

Vgs and Vds command log

Temperature

ADC6242 + Type-K probe

Die temperature, actual & set

Dose

X-ray supervisor + HP 4140B diode

Cumulative TID exposure

Phase markers

Experiment orchestrator

Setup · stress · anneal · characterise

Recording & performance

400 kHz
Acquisition rate
200 kHz
Analog bandwidth
~40 nA
Noise floor
10 µA · 16-bit
Dynamic range
2.5 µs/pt
Time resolution
> 1 TB / run
Dataset size
< 1 s
Full-timeline latency
< 5 s
Detail-zoom latency
~4 h (4-core)
Cache build
> 2 days
Recording time

Browsing the whole campaign at full resolution turned up something the usual workflows average away: discrete charge-trapping and detrapping steps, caught while the dose was still building up. Because every instrument shares one clock, I could check each step against the live bias, temperature and dose and rule out drift, handling and ESD — what is left is the device's own traps switching. That is the result our RADECS 2026 paper is built on, and it is very hard to see without continuous, in-situ acquisition.

RADECS 2026 paper →Bachelor thesis →MESS symposium
research.yurii.ch/l/mess