March 2, 2020
HistoryZeptobars started in 2012, when I was working on "homemade" ICs. Before trying to design or make your own it is beneficial to look at work of others. This on-hands experience is invaluable. Original goals were to observe manufacturing of MIFARE tags at Mikron, which then was starting manufacturing on 180nm technology (this was concluded in 2013), look closer at KR580VM80A - CPU of my first computer (this was concluded with full reverse engineering in 2015) and finally - look with greater details at simpler CMOS chips (7400/4000 series) as a starting point for DIY design.
Originally I got few "handheld" microscopes - one Soviet one and 2 Chinese. It soon became very clear that one cannot see anything with it let alone photograph. So I bite the bullet and purchased Chinese metallographic microscope BM-158J which cost me 1600$. If only I could go back in time and not buy dark-field, 100x dry lens and DIC attachment... Quality of microscope was acceptable for visual use, but was insufficient for photography. If you inspect early photos from 2012-2013 in 100% magnification - you will often see blurred areas, chromatic aberration and noise.
Over the next 7 years microscope was upgraded with gradual quality improvements:
- Original 10x/0.25 lens (the most useful) was misaligned, and I ordered another copy from manufacturer which was aligned ok.
- I was able to find adapters to RMS and M26 lenses for this microscope, and bought my (currently) favorite Olympus UMPlanFl 10x/0.3 lens, which is optically perfect (was marked as "phase shift").
- Original trinocular head was replaced with original Olympus U-TLU tube lens with custom adapters which increased quality once again.
- Many 0.5x camera adapters tried, including original Olympus U-TV0.5XC-3. This again improved quality, but still 0.5x adapter was a source of significant quality degradation.
- After trying many cameras with larger sensors, with the help of Patrons ToupTek E3ISPM20000KPA was bought. It's 1" sensor allows to use passive 1x camera adapter which significantly increased image quality. I can say that this gave the largest quality improvement over the years.
- More specialized and higher quality lenses were bought: Olympus 100x/1.35, 50x/0.8, 10x with water immersion, UPlanSApo 4x/0.16, lenses with glass thickness correction (20x/0.45 and 50x/0.70 LCD).
- Lateral chromatic aberrations corrected with tca_correct (part of Hugin). Calibration is semi-automated.
- Tungsten illumination source replaced with RGB LED (not white). It improved image quality after tca_correct as each color plane is sharper due to narrower illumination bandwidth. Also, colors are somewhat more saturated which is beneficial for IC photography.
- Automatic metalization etcher & solution developed - 2:4:1:1 solution of acetic acid(70%):H2O:HCl:HF(40%), 300 seconds etch at 80°C. Gives reliable results on aluminum chips, while copper still requires some luck.
- Finally, in 2019 main microscope was upgraded to used Olympus BX60. Even being used, it was still a major investment. Optical quality was perfect from the start this time . It fixed all remaining optical issues that were remaining in the illumination path (the only optical part which was original in old microscope). The only thing I needed to do was to buy a neutral illumination cube - all stock cubes were for fluorescence. Eventually it will also be upgraded to RGB illumination.
FutureThere is still quite a number of things in the pipeline for the following years:
- Olympus RGB illumination
- More automation - motorized panoramas, auto-stitching/rotation/cropping
- Switch to monochrome camera with more wavelength options (UVA to NIR)
- Laser decapsulation
- Diffusion ROM reading
- Mechanical de-layering
- Ion de-layering
- Scanning electron microscope
- Transistor cross-sections (mechanical/ion)
- Microfocus X-Ray for complex plastic assemblies and microoptics
- Help with a script or contribute to Hugin to allow pre-calibrated chromatic aberration correction done during panorama stitching. This will allow to have only 1 image re-sampling vs 2 we have now (chromatic correction using tca_correct + panorama stitching with geometric distortion correction and rotation).
- Contribute to Hugin to improve control points detection on regular structures (like DRAM/SRAM arrays), where control points should only be detected in non-regular parts.
- Contribute to Hugin to support "sequential XY" control points detection, where global coordinates of each 2 consecutive frames are translated by X or Y, but not both. X/Y vector common for all frames to be detected (basically, angle between XY stage and camera). This will greatly improve CP detection on regular structures with low overlap.
- Contribute to Hugin to improve multi-threading in cpfind (it's not always using all cores on >4 core machines on all stages of processing).
- Contribute to Hugin to improve final panorama stitching speed for large projects, improve CPU utilization on large CPUs (25+ frames). Possible "RAM vs speed" trade-off.
- Ultimate goal for reverse engineering is Intel 386DX, which is a huge task even with full automation.
Software flow improvements, mainly Hugin-related - where you can help us:
Who is behind Zeptobars?My name is Mikhail Svarichevsky (LinkedIn). I liked electronics and microelectronics since childhood, yet never had an opportunity to work in the field: I worked in software for many years, and currently - CTO of Wayray, where I am involved in optics, holography, optoelectronics (lasers and integrated photonics), engineering and embedded software.
Copyright notesAll die-shots published on this site are licensed under a Creative Commons Attribution 3.0 Unported License.
You are allowed to use, publish, translate or repost our articles and die shots in electronic and printed media provided that you include link to the source on our website.