SCIENTIFIC INSTRUMENTS / LATEST COVERAGE
Berkeley Lab brings live analysis to the electron microscope.
A new camera and a direct route to graphics processors let scientists examine a torrent of microscope data and adjust the experiment while it’s running.
Strong and faint signals, in one shot.
An electron microscope can tell you more than what a material looks like. By measuring how much energy electrons lose as they pass through a sample, researchers can learn which elements are present and how atoms are bonded. It’s called electron energy-loss spectroscopy. Capturing the strongest and faintest signals traditionally takes two carefully timed measurements that have to be lined up afterward.
A camera developed at Lawrence Berkeley National Laboratory’s Molecular Foundry captures both at once. In a November 19 technical report, NVIDIA describes how the two teams handled the resulting stream of more than 500 gigabits per second. That’s the rate of incoming detector data in this setup, not a measure of how fast every microscope can run.
From the sensor to the screen.
Instead of writing the raw data to disk first, the team sent it over Ethernet directly into the memory of graphics processing units (GPUs), using NVIDIA’s DAQIRI data-acquisition library. The processors assembled and corrected the images and turned them into spectra while the camera was still collecting data. NVIDIA reports that researchers could inspect the results and adjust their settings or region of interest during the measurement.
NVIDIA’s November software release adds options for handling time-sensitive data, including letting compatible network cards put packets back in order before the software has to do it. The release notes describe tools for instrument builders; the microscope example shows why this kind of data plumbing is worth getting right.
A shorter feedback loop.
Instruments like this are an opening for Super Intelligence (SI) in the lab. If a system can read a measurement while it’s happening, builders can explore tools that help scientists spot an unexpected signal or decide where to look next. The microscope example shows live processing and researcher-directed adjustments, not an autonomous experiment. The advance is that the next decision can happen while the specimen is still under the lens.