4nm 3D chip X-ray sets world record
Cette publication existe aussi en Français
Researchers in Switzerland have achieved a record resolution of 4nm for a 3D X-ray of an AMD CPU.
The team at Paul Scherrer Institute (PSI) used X-rays from the Swiss Light Source SLS at PSI. Instead of using lenses, the team used a technique called ptychography, where many individual images are combined to create a single, high-resolution picture.
The team worked with EPFL Lausanne, ETH Zurich and the University of Southern California on the X-ray images with shorter exposure times and an optimised algorithm.
While scanning electron microscopes have a resolution of a few nanometres and are therefore well suited to imaging the tiny transistors and metal interconnects that make up circuits, they can only produce two-dimensional images of the surface.
AI improves 3D X-ray imaging for package failure analysis
“The electrons don’t travel far enough into the material,” said Mirko Holler, a physicist at SLS. “To construct three-dimensional images with this technique, the chip has to be examined layer by layer, removing individual layers at the nanometre level – a very complex and delicate process which also destroys the chip.”
However, three-dimensional and non-destructive images can be produced using X-ray tomography, because X-rays can penetrate materials much further. This procedure is similar to a CT scan in a hospital. The sample is rotated and X-rayed from different angles. The way the radiation is absorbed and scattered varies, depending on the internal structure of the sample. A detector records the light leaving the sample and an algorithm reconstructs the final 3D image from it.
“Here we have a problem with the resolution,” said Holler. “None of the X-ray lenses currently available can focus this radiation in a way that allows such tiny structures to be resolved. Our sample is moved such that the beam follows a precisely defined grid – like a sieve. At each point along the grid, a diffraction pattern is recorded.”
The distance between the individual grid points is less than the diameter of the beam, so the imaged areas overlap. This produces enough information to reconstruct the sample image at a high resolution with the help of an algorithm. The reconstruction process is rather like using a virtual lens.
The sample, an extract from a commercial AMD processor, was supported by the gold-coloured pin in the centre of the picture. Less than 0.000 005 metres in diameter (about 20 times smaller than the width of a human hair), it was cut out of the chip using a focused ion beam and placed on the pin.
As the beam is wobbling, each image changes slightly, and these changes are used to track the actual position of the beam caused by the unknown vibrations. The algorithm compares the positions of the beam in the individual images. If the positions are the same, they are put in the same group and added to the sum.
By grouping the low-exposure images, the information content can be increased. As a result, the researchers are able to reconstruct a sharp image with a high light content using the flood of short-exposure pictures.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
