http://zeptobars.com/en/ Microelectronics. Die-shots. Artificial intelligence. Lasers. en-us Tue, 10 Jun 2006 04:00:00 GMT Sun, 24 Sep 23 12:18:46 +0300 webmaster@zeptobars.com 120 10 http://zeptobars.com/en/read/Motorola-MC14490-CMOS-button-debouncer-Bounce-Eliminator Die size 2400x2048 µm.


]]> Fri, 01 Sep 23 09:13:15 +0300 http://zeptobars.com/en/read/TimesVano-G170DM-transient-overvoltage-protection-SLIC TimesVano G170DM can be found in wired phone line protection circuit. Can survive 350A current for 2µs, clamps both positive and negative voltage spikes.
Die size 3148x2292 µm.


]]> Thu, 01 Jun 23 11:21:48 +0300 http://zeptobars.com/en/read/TL432-TL431-Precision-Programmable-Reference-shunt-regulator TL431 is pinout.
Die size 810x662 µm.


]]> Mon, 29 May 23 12:35:27 +0300 http://zeptobars.com/en/read/RCA-CD4035A-CMOS-4-Stage-Parallel-in-out-Shift-Register Die size 1733x1804 µm.


]]> Mon, 01 May 23 10:42:07 +0300 http://zeptobars.com/en/read/zeptolab-upgrade-2 support of Zeptobars patrons - several vital hardware upgrades finally happened which will increase image quality and will make work more streamlined : custom wavelength RGB illuminator and 2-axis goniometer. Finally, we've also got Mastodon - @zeptobars@techhub.social - let's keep in touch there (just in case).


More details below:

First is custom RGB light source. I had to admit I've spent way too much on it due to fine-tuned wavelengths of the diodes.
My previous RGB light source that I used with old Chinese microscope used regular 450-520-620nm RGB leds. These do not match Bayer filter matrix in Sony imaging sensors very well (colors bleed into wrong channels reducing saturation and reducing efficiency of software correction of lateral chromatic aberration). So I decided to go for 440-545-660nm, which matches RGB filters and covers wider spectral range.

440nm is shortest wavelength that can be used on fluoride microscope lenses without significant quality degradation (they are designed for ~436nm and longer). Getting 440nm LED was straightforward - as it's just binning of commercially available GaN diodes. 5 batches from different suppliers was enough to nail down the wavelength.

660nm is the easiest. My very first batch was spot-on.

545nm diodes though are very challenging as they are in "green hole", where all LED semiconductors shows poorest quantum efficiency. In blue direction GaN starts working very well around 510-520nm. In yellow/red - multiple options starting at 580. I've tried to find extremely binned GaN diodes, and maybe 10 different batches gave me 530nm maximum. Phosphor-converted green LEDs have extremely wide spectrum, and definitely not suitable for this application. Finally, I've ordered GaP diodes from China, and in 4 batches one was perfect at 545nm. Surely, quantum efficiency is much lower, but I don't illuminate whole room anyways.

Per-channel adjustable current allows to set white balance to make sure gains of all RGB channels are 1.00 => lowest noise on the image. All extra capacitors (and inductors) - form CLC filter to ensure supply ripple does not introduce noise on progressive scan camera sensor.

LED light is side-coupled into 3mm optical fiber, which goes into microscope. I still want to build second version with the same LEDs with higher coupling efficiency. Otherwise I am already quite happy with it.

RGB illumination have 3 main benefits:
1) We work only on wavelength where lens shows highest quality,
2) RGB color channels are "independent", and correction of lateral chromatic aberration is much more efficient,
3) Color saturation is much higher without any special post-processing. See example below, exactly the same chip, similar post-processing but different light source:

White LED:

RGB LED:

2-axis goniometer:
On the second upgrade - while ~80% of chips are flat enough to be photographed on regular XY stage, remaining ~20% are slightly tilted and became royal pain in the ass (typically in metal or metal-ceramic packages, due to uneven soldering or cases which impossible to lay flat). Dual-axis goniometer with micrometer adjustments from "ZY Automation" finally solves this issue.

You can also read previous article about how things worked in our lab in the past. ]]> Mon, 06 Mar 23 02:50:12 +0300 http://zeptobars.com/en/read/National-LM3915N-1-Dot-Bar-Display-Driver-thermometer-logarithmic Die size 2394x2545 µm.


]]> Wed, 01 Mar 23 13:45:50 +0300 http://zeptobars.com/en/read/Shenzhen-Fuman-LM358S-general-purpose-opamp one from Ti.


Not sure manufacturer has a website. "Founded in 2001, Shenzhen Fuman Electronics Group Co., Ltd. is a state-level high-tech enterprise engaged in the research and development, packaging, testing and sales of high-performance analog and digital-analog integrated circuits. Currently has more than 400 kinds of IC products in the field of consumption including power management, LED driver, MOSFET and so on."


]]> Wed, 01 Feb 23 13:39:41 +0300 http://zeptobars.com/en/read/Toshiba-2SK369-low-noise-JFET BF862 we see lots and lots individual transistors in parallel.
Die size 765x765 µm.


]]> Mon, 16 Jan 23 00:38:30 +0300 http://zeptobars.com/en/read/Casio-F-91W-OKI-quartz-watch SensorWatch which is built around F-91W - I was curious to see how original chip looked like.

Surprisingly, chip is more complicated than I anticipated and digital part is less than ~50% of the die area. F-91W has trim jumpers on the PCB to adjust part-to-part variability of the quartz crystal. Looking at die complexity - it might be that they use capacitive tuning of frequency rather than skipping/adding ticks. Also, it might happen that they even have coarse/cost-efficient temperature compensation. Casio specify accuracy of 30 seconds per month, but some F-91W watches are within 6 seconds per month which requires good adjustment & luck for non-temperature compensated watch.

Latest versions of the watch work 7-10 years on a single battery, while early versions lasted "only" 2 years. While original F-91W was introduced in 1989, this die is designed in 2006. If someone has original version from 1989/early 1990's - it would be interesting to compare if it's a "die shrink" or re-design with improved performance.

You can compare die complexity to Luch quartz watch. It's not a digital watch, but see how much less analog circuitry was there.

Die size 2990x2541 µm.



Pinout suggested by Joey Castillo:

]]> Sun, 01 Jan 23 12:45:13 +0300 http://zeptobars.com/en/read/RCA-CA3031-opamp
Die size 1142x1134 µm.


Schematic below:

]]> Sun, 01 Jan 23 00:29:21 +0300