Hauling 2000lbs of scientific surplus across an ocean on a quest to image my red blood cells at home
Every nerd remembers the first time they saw a scanning electron microscope (SEM) image of a fly’s compound eyes, or the scales of a human hair:
As a newly-graduated engineer, I stumbled on an eBay listing for a vintage AMRAY 1000 SEM and impulsively made an offer which was (luckily, in retrospect!) declined.
I lived in my parents’ house with no attached garage, in a different country, and lacked the knowledge (and means) to import, install, repair, and operate such an advanced device. I promised myself that one day when I had all of these things, I would get an SEM.
That day finally came.
My goal? An SEM image of my red blood cells, taken in my garage.
In theory this should have been a straightforward (but tedious) process:
In practice, there were complications.
Before searching for a surplus instrument, I defined my requirements.
Since I would not have a manufacturer service contract, ease of maintenance and repair was at the top of the list: accessible schematics, available parts and consumables, and documented proof (i.e. Youtube videos, blog posts) of others restoring the same or similar model.
The primary consumable for an SEM is the electron source. Although a Field Emission SEM (FESEM) offers higher resolution, they use expensive tips (~$1000 each) versus the simple hairpin tungsten filaments in a traditional SEM (~$60 each). I assumed that I would break at least one electron source along the way (through ignorance, brute force, or natural causes).
Tungsten filaments are simple enough that I may even make my own. I limited the search to tungsten-filament SEMs.
For access to schematics and spare parts, brand popularity helps. On the secondary market, one brand dominates: JEOL. JEOL has a long lineage of reliable SEMs, with many shared design and components throughout their product lines.
Many models use the same column and electronics design, and there are scanned schematics and manuals scattered across the internet for various models. One of my favorite home-lab YouTubers chose JEOL for his garage SEM and documented it well. Other home-lab JEOL users can be found on YouTube. I tried to limit myself to JEOL-branded SEMs, but would reconsider for the right deal.
Digital image capture was another requirement. Early SEMs displayed images on a main CRT and sometimes recorded to film (either from the main CRT or from a dedicated smaller, higher resolution one). By the mid-80s, digital SEMs were created, storing the image digitally in a buffer.
It would be possible to convert an older analog device to digital capture, but it would take a lot of work - I would need to design electronics to control the electron beam sweeping in sync with an analog to digital image capture solution. An even more basic solution would be to use a digital camera to capture the displayed CRT image, but quality would suffer. I decided to find a SEM where the manufacturer already provided digital image capture.
In 1997, JEOL released their first PC-controlled SEM (the JSM-5600):
Prior to PC-controlled SEMs, all beam control, vacuum control, and image capturing logic was implemented in hardware in a large control console. With the PC-controlled SEMs, much of this logic was moved to an application on the PC. The control console is still present, but is significantly smaller and less complicated. The drawback of this approach is that without the software, there is no way to operate the device. If I bought a PC-controlled SEM, I needed the software.
There is another type of SEM which I eliminated in my search - the benchtop SEM. These are smaller units that operate at a lower voltage but otherwise operate the same as the larger full-size SEMs. Two common models that appear on the secondary market are the JEOL NeoScope and the Phenom
The primary disadvantages of benchtop SEMs are reduced specimen size, lower accelerating voltages (less penetration), and reduced magnification. I watched for a good price on a benchtop SEM, but my preference was for a full-sized unit.
I started by contacting local universities. Each university has multiple “core facilities” that manage equipment and budgets independently. Researchers get access to these shared resources for their own work. I found one core that was upgrading their SEM, but their old one had already been claimed.
I tried medical research centers, but no luck. Then surplus auctions. After months of watching auctions nationwide, no SEM met my needs. A few transmission electron microscopes (TEMs) showed up, and I considered bidding - but I had no good reason to own a TEM. They are significantly more dangerous (higher accelerating voltage, more risk of X-ray exposure) and require a ultramicrotome for sample preparation (slicing samples into sections 50nm thick!) - a tool that is itself hard-to-find. For all this trouble I would be rewarded with the ability to image individual atoms, and who doesn’t want the ability to see atoms in their garage? A garage TEM may (read: will) be in my future. (I have since learned about atomic force microscopy, and plan to build my own AFM)
To get a suitable SEM, I needed to increase my risk tolerance and turn to the last resort: used equipment brokers. I learned the hard way why they are known as “scrap metal dealers”.
After months of going back and forth with used equipment brokers, a viable option appeared: a JEOL JSM-5600LV located in Europe. The seller claimed it was in working condition prior to decommissioning, and it included with a vacuum pump, spare filaments, the control console, and more.
Before committing, I needed confirmation: -was a PC included? -was the JEOL software included? -how many spare filaments were included?
This 1999-vintage SEM runs on obscure JEOL software for Windows NT 4.0/XP. Without that software all I’d have is 2000 pounds of metal taking up half my garage. In addition to the software, specific PCI cards are required to interface the PC and the SEM. I wanted those as well.
The spare filaments were not critical, but at ~$60 each it would be nice to receive a large supply.
A few weeks later I had written confirmation that the correct PC and 48 spare filaments were included. I wired the money and was transferred to a settlement specialist.
The device was sitting in storage at the seller’s facility, and the broker asked for clear instructions on how to secure everything for crating. With help from various tweets, YouTube videos, eBay listings, and ChatGPT, I pretended to be an SEM-shipping expert . The key points: the electron column must not move, the specimen chamber should remain closed, the control arms should be protected, and everything should be wrapped as tightly as possible.
I summarized these requirements and provided example images:
A week later the broker had 1800 pounds of “scrap metal” spread across 3 crates, ready for shipping.
The crates were in Scotland, and I was in Texas. A container ship was the only way they’d get to me at a reasonable cost. The “One Triton”, in particular:
Thanks to the wonders of AIS I was able to track the crates on their journey across the ocean. The crates were expected to take 4-6 weeks to arrive at the Port of Houston.
Before the crates could be on their way, I had to learn how to import an enormous piece of typically-quite-expensive, special-purpose industrial equipment into the United States.
Apparently I needed an “ISF and import customs clearance agent” to proceed. As usual, ChatGPT made me a quick “expert”:
I got in touch with a local customs broker and completed a Power of Attorney. I received a Packing List and Bill of Lading from the seller to sign off on:
The broker filed the ISF form and the shipment was cleared. One section of the US customs documentation was interesting: because of the potential for this device to be modified for use in the manufacture of semiconductors, I had to agree that I would not export it to any of the “bad guys” for that purpose:
The duty, taxes, and filing fees were not too bad - the biggest expense was a “customs ISF bond”. Once again, ChatGPT came to the rescue:
If I was planning to import multiple containers throughout the course of a year, I would have arranged for a continuous (annual) bond which applies for all importations for 12 months for a flat fee. In this case I paid the one-time fee and awaited delivery.
It’s always fun tracking the progress of a shipment you’re excited for. It is even more fun when it is 1800lbs of machinery on a boat somewhere in the Atlantic. Things were going smoothly until two back-to-back incidents threw a wrench in the works.
The first: Hurricane Milton, the second-most intense hurricane to ever hit the Gulf of Mexico - just as the SEM was in the neighborhood:
Things looked sketchy for a few days, but luckily Milton weakened as it approached the Florida Coast and the ship was back on the move - until the Houston Port strike began:
Things again looked sketchy for a few days, but the strike ended after 3 days. The ship finished its journey and the crates were finally in the US:
Now I had to deal with final delivery. The terms of delivery were “kerbside”. This means the trucking company was only obligated to drop the crates at the end of my driveway. That would have been a problem, as I don’t have an easy way to move 1800lbs up an inclined driveway, over a lip, and into my garage. I could ask (read: beg) the delivery person to make the extra effort, but I couldn’t count on that succeeding. I considered purchasing a pallet jack, but none were available at a fair price. Since rental pallet jacks are available I left that decision for the day of delivery. I purchased heavy duty tarps and straps to prepare for the worst case scenario (the crates being left curbside on a rainy day).
On delivery day, I built a makeshift cardboard “ramp” to deal with the lip leading into the garage (thanks to a pro-tip from an experienced friend):
The delivery guy was kind enough to finish the job with an electric pallet jack:
PRO-TIP: Consider the maximum height of your garage door! I realized on delivery day I had never checked if I had sufficient clearance. It was a close fit, but there was enough room.
Everything fit in the garage with room to spare:
At last, I had my SEM.
Next came the fun part, unboxing.
The custom-built crates were both nailed and screwed shut. A couple of hours with a drill, nail-puller, and pry-bar and I was in.
So far so good, everything was well-packaged and sealed tightly against moisture. I wanted to be able to easily transport the microscope in the future, so I kept the tallest crate as intact as possible. The other crates would not be needed so I was more aggressive when opening them.
I finally got a look at what I had. Everything seemed to be present. The electron gun was not fully closed, it appeared a set screw was broken at some point. This meant the electron column was not under vacuum. The specimen chamber remained under vacuum.
There was a broken adapter and BNC on the control console and it clearly has been bumped at some point prior to crating:
The shock detector and tilt detector were not triggered during transport, which was a good sign:
The provided vacuum pump still had oil in it, and the oil looked good. No high vacuum hoses were included:
The foreline trap (prevents oil backstreaming from the rotary vane pump into the SEM vacuum system) looked good:
The PC had no hard drive, this was a big problem. It also had no capture card or SCSI card, which was another big problem. The LCD monitor was damaged, but still worked.
I received a box of spare parts which could come in handy in the future. The filaments were a serious problem - only a fraction of the promised amount were provided and almost all of them were broken.
I also received a few binders of documentation.
Before going further I needed to deal with the missing and broken filaments.
Long story short, the broker stopped communicating once I notified them they were short 33 filaments, with 9 others damaged. At a replacement cost of $60, that’s $2520 worth of supplies I was promised but wasn’t delivered. End result: complaints filed with the FTC, the Texas Attorney General, the BBB, and more. Ultimately no resolution (yet).
PRO-TIP: Don’t buy from scrap metal dealers. Being patient with auctions or local sellers is the better way to go.
I dug through my documentation and found the initials of a person at the original company who owned the device. After a bit of sleuthing I got in contact with him. He located a few of the filaments and provided additional information on the device. Most importantly he confirmed that the device was 100% functional at the time of decommissioning.
I started with a superficial cleaning of the exterior. Along the way I replaced any missing screws, nuts, and other small hardware.
I went piece by piece for the refurbishment.
The provided PC was in rough shape, and was missing a hard drive and the required PCI cards. I started by installing an SSD and maxing out the RAM. The latest operating system supported by the JEOL software for this device was Windows XP, so I started with that. Unfortunately, I had no success installing the operating system:
After many hours of troubleshooting (boot disks, boot CDs, swapping the SSD with an HDD, Windows NT4, Windows 2000, replacing cables, power supply, and more), I remembered the capacitor plague. As a 2000-vintage PC, the motherboard may have been affected. After close inspection, a few bulging caps were present. Replacing all of the capacitors got me back on track. I was rewarded with Windows XP bliss.jpg in its natural habitat.
Now that I had a working computer, I could track down the missing components.
First of all I needed the appropriate SCSI card to interface with the control console, which had a 50-pin SCSI connector on it. After a bit of sleuthing I found that an Adaptec PCI card is what came with the system originally. I tracked one down on eBay:
The capture card was a bit harder to figure out. I could see from the label on the back of the computer that the system was using a Matrox Orion PCI capture card with BNC video, so I initially tried to replace it with an identical model - but they were very expensive on the secondary market:
Further research revealed at least two other PCI capture cards used by JEOL SEMs in the late 90s: the Euresys Picolo series and the Integral Technologies FlashPoint series. The fact that I had evidence of the use of commercially-available capture cards from three different manufacturers meant that I should be able to substitute with any PCI capture card that meets the required specifications. But what were those specifications?
A fellow vintage-JEOL-owner reported to me that the video signal from these control consoles is not standard NTSC/PAL: the timing is unusual, the signal has vsync and hsync pulses, and the resolution is variable. He reported that he had seen the same type of signal on JEOL instruments going back into the 80s. This lack of NTSC/PAL ruled out the use of extremely common (and cheap) TV-tuner PCI cards.
Looking at the back of the control console, I saw two possible video connections:
The “stock” configuration was an S-Video port. But my device also had the MP-65250 (ESITF) add-on card installed, with Video Out via BNC. ESITF likely stands for “external scan interface”, as this card allows external control of the scanning coils for use with an EDX system (for performing elemental analysis on a sample using x-rays). The coils are controlled via the serial ports, and the output from the detector comes through the BNC. In my device it seems like this was the video port that was being used, so I looked for a PCI capture card offering BNC video.
I still didn’t know what the format of the video was. I did know the video would be monochrome, since the signal coming out of the microscope is just an intensity value from the photomultiplier tube (or similar sensor). By comparing specifications of the three known-good capture cards, I noticed something unusual they had in common: support for RS-170 video signals. This was a monochrome video format that was never found on any consumer-level PCI capture cards, so the odds were very good that this is what is being used by JEOL. More digging on the internet revealed a university project from 1995 to add networking functionality to a JEOL SEM - it seemed to confirm my suspicion: “The microscope image is carried to a television monitor at each remote station using an RS-170 video feed from the JEOL 6100 provided by the manufacturers”.
I tracked down a Euresys Picolo PCI capture card with BNC input and RS-170 support:
The final missing component was the toughest - I did not have the late 90s JEOL software which would make this machine more than a pile of scrap metal. Scouring the internet revealed images of what the disks and floppies containing the software looked like:
So I had a target. While searching I learned about the JEOL JSM-5610LV which is an identical model to the 5600LV in every way, with the only change being a software upgrade with Windows XP support. This expanded my opportunities to locate the missing software. Emails to various owners at labs, eBay sellers, and elsewhere did not result in anything. But with luck and perspiration, I ended up finding an FTP site hosting a version of the 5610LV software. It installed successfully:
The floppies were still missing (according to the documentation I had, two floppies unlocked features in the main JEOL software to support the addon cards), but this was all that was needed to get started.
I received one rotary vane vacuum pump with the system. Unfortunately as a device with a low-vacuum mode (hence the LV in 5600LV), I needed two. The documentation provides a diagram of the vacuum system:
The first rotary vane pump (RP) is used for roughing the oil diffusion pump (DP) and the rest of the microscope vacuum system, the other rotary vane pump (RP2) is used for holding the specimen chamber at a different vacuum level from the rest of the system when operating in low-vacuum (LV) mode. (I will discuss the purpose of LV mode when I put it to use).
Before looking for a second rotary vane pump, I looked up the specifications of the pump I have:
Leybold D65BCS
Pumping speed: ~1075 L/min
Ultimate partial pressure: 0.01 Pa
And compared this to the specifications of the “stock” pump that JEOL typically provided with the JSM-5600LV:
ULVAC G-100
Pumping speed: 100 L/min
Ultimate partial pressure: 0.07 Pa
The Leybold is seriously overspecified for this purpose. A quick google revealed that it is no ordinary pump:
That made up for the lost filaments. I decided to keept the Leybold and put it to good use, but this introduced a new problem - the Leybold requires 240V, 3-phase power, at a continuous 2200W. I dealt with the power situation later, but I obviously did not need (or want!) the second rotary vane pump to be this ridiculously overpowered. Scouring the surplus market I found a suitable replacement pump, the ULVAC GLD-136C:
Next I inspected the JEOL’s built-in oil diffusion pump:
Everything looked fine, the watercooling baffle was in good shape and the oil level in the high voltage tank was good. Any problems would be revealed during initial pump-down. (To understand the ingenious way an oil diffusion pump achieves extremely low vacuum, watch this video. To see the process in action, watch this video of a transparent glass version.)
Next I worked on the vacuum connections. The SEM and the foreline trap both have a tapered flange with a diameter of 24mm on the side that goes to the rotary pumps:
The foreline trap is connected to the SEM with metal vacuum hose:
The Leybold pump uses KF40 quick-release flanges:
And the ULVAC uses a KF25 flange:
Proper vacuum hoses are expensive, the ones that originally came with the system are thick rubber:
Luckily a cheaper solution was available: thanks to the experimentation of a fellow vintage-JEOL owner, I had proof that steel-reinforced PVC tubing secured with hose clamps holds up to the level of vacuum required:
Looking at the documentation, there should have been a vibration isolator/dampener between our rotary vane pumps and the microscope (to keep vibrations in the specimen chamber as low as possible for maximum image clarity/resolving power). Research indicated that this vibration isolator is simply a very heavy weight:
I could make my own using concrete mix and a large bucket but decided to wait until I determined if it was necessary (e.g. if vibration is reducing the image resolution/quality).
I needed to quantify how good the vacuum pumps were functioning - were they able to achieve at least 0.07 Pa? This was within the spec of both of the vacuums, but I wanted to verify. A cheap and “good enough” solution was to use an HVAC micron gauge. I used a CPS VG200 with some adaptors:
TODO: photo and numbers
The ULVAC reached X.XXX Pa, the Leybold (details below on how it was powered) reached X.XXX Pa.
TODO: photo
The column of the device contains a number of o-rings. I assumed most were still working, but replaced the one at the top of the electron gun which had been exposed to atmosphere during storage.
TODO: photo
There is a lot of debate regarding the use of high vacuum grease when working with o-rings, but the JEOL documentation advised us to use it. So I acquired a small amount of Apiezon M. Other greases would have worked (e.g. molykote high vacuum), but I wanted as contaminant-free a vacuum as possible, so why not use the best?
Everything looked good with the vacuum system at this point, I planned to replace the electron gun o-ring when I got to doing a deep-clean of the device prior to first power-on.
TODO: photo
The diffusion pump gets extremely hot during operation - coolant must be flowing through the cooling coils at all times. The documentation specified a flow rate of 2L/min, maintaining the coolant temperature at 20C(+/-5C). One quick-and-dirty idea was to run a constant stream of cold water from my residential water hose, draining it out onto the lawn. This would work in the short term, but a long-term solution was used instead: a dedicated industrial chiller. Research indicated that the go-to cooler for this purpose is the generic CW-5200 chiller. It can achieve the temperature and flow rate specifications and is relatively inexpensive.
Since things are never easy, the first one arrived damaged. The subsequent replacement worked fine, however. I hooked it up with PVC tubing.
And swapped in matching barbs on the diffusion pump watercooling baffle:
One thing I had to consider is that the garage this SEM is housed in will experience cold temperatures for a few weeks of the year. Although it is unlikely to reach freezing, I still needed to be prepared for that. In addition, I needed a suitable biocide to prevent growth in the water cooling lines. A mixture of 25% Propylene Glycol to 75% distilled water achieved the goal. It is not a rust inhibitor (potentially needed since I have mixed metals in the cooling loop), but it should be fine.
As a Japanese device, the SEM runs off of 100V whereas in the USA I am provided with 120V. This step-down conversion was easy to achieve with a variable transformer - I put one together with a display to monitor the voltage:
The Leybold vacuum pump presented an interesting problem: it needed 220V of three-phase power. I had access to single-phase 240V via a NEMA L14-30P socket used for an electric clothes drier, however it was located 50ft away from the SEM. The socket can handle a maximum of 30A of current, which meant I needed 50ft of 10 gauge extension cable. This is a common specification for power generator extension cables, however I needed an adaptor for the more common NEMA L14-30R the extension cable used.
I now had 30A of single-phase 240V in the right location, but I needed three-phase 220V. With a sufficiently beefy variable-frequency drive (VFD), I was able to take that 240V single phase and output 220V three-phase. The Leybold vacuum is 3HP, running on 220V at 10A (2200W), so I overshot with the VFD to be safe. I wired it up with a NEMA L13-30R connector to fit the extension cable:
I now had all the appropriate voltages, but anticipated problems with current draw when the entire system was powered on. I dealt with that problem later.
It was finally time to wire everything up. One unknown was whether all of the cabling between the microscope and control console was intact and present. I had been told that the device was properly decommissioned, but there are many SEMs on the surplus market where the uninstaller took a destructive shortcut and cut the cables. This would have been a nightmare to repair. Luckily the cables were not cut in my unit, and most had a convenient labeling system applied:
It was tedious work, but I connected everything where it belongs:
For cable management, there were a few JEOL-provided cable stays provided and I printed additional ones to tidy everything up:
The device was coming together nicely:
TODO: photo
It was almost ready for first power-on. Before taking that risk, I took some steps to protect the device.
I inspected all of the PCBs in the microscope and the control console, and found it uses many EEPROMs, microcontrollers, and PLDs. These chips are all fairly standard and can be easily replaced, but the code contained within them can’t be. If any of these chips were to be damaged or destroyed, I would be in a tough spot. So for extra insurance, I took the opportunity to make backups.
A universal programmer was able to back up most chips, expect for an unusual Hitachi H8/300 microcontroller in a PLCC-84 package which JEOL used on multiple boards:
These chips needed an adapter in order to be dumped. Luckily, a fellow vintage-JEOL-owner has designed one. I had some PCBs made, purchased the necessary sockets, and soldered one together:
TODO: photo
Dumping the chips was tedious work, but was worth the effort for the insurance.
TODO: photo
Contaminants in the vacuum system of an SEM impair resolution and performance. Since my device had been in storage for an unknown amount of time (with the electron column exposed to atmosphere), a deep clean of the electron column and other vacuum system components was needed. A deep dive into how to clean high vacuum devices and other industrial/scientific equipment led to the purchase of the following chemicals:
I didn’t want lint left behind on any surfaces after cleaning, so I used lint-free materials:
With vacuum devices even the tiniest scratch or groove can cause significant problems with maintaining seals, so I used plastic tweezers when replacing o-rings or performing any operation on surfaces which will be under vacuum:
I cleaned every interior surface using Everclear:
TODO: photo
I ran some metal parts through an ultrasonic cleaner with the Alconox solution:
TODO: photo
I cleaned the o-rings with Everclear, and before re-inserting them I applied a very thin layer of Apiezon M high vacuum grease (following the procedure outlined by JEOL):
The goal was to make the o-ring surface shiny, to fill in the rubber pores:
TODO: photo
For the vacuum hoses, I wet a clean-room wipe with Everclear and pulled the wipe through the length of the hoses. Before clamping down the vacuum hoses, I cleaned the flanges and applied a very thin layer of Apiezon M high vacuum grease:
TODO: photo
As the specimen chamber was still under vacuum, I would need to release the vacuum before inspecting. This could only (safely) be done using the controls on the console, so it had to wait until after initial power-up.
TODO: power it up
TODO: release specimen chamber vacuum