41J Blog

This post is a reply to Richard Katz post on “How to Restore Japan as a startup nation”.

I have a lot of thoughts on this, having worked for a number of US, UK, and Japanese startups in various positions (up to CTO and founder). I’ve also created companies in US, UK and Japan. And received Angel funding for my US company (Reticula).

In this post I’m going to compare the startup environment I’ve seen in Japan and the US…

Company Creation

Summary: Company creation in Japan takes months and >$3000. As opposed to days and $10s in the US.

Let’s just briefly talk about the most basic part of the process, creating the company. I’ve created companies in the US, UK, Estonia, Hong Kong and Japan. Japan is by far the most complicated and expensive company creation process.

To create a Japanese company you need to:

• Pay ~$2000 in government fees.
• Create a custom set of 3 ink stamps (Hanko) for the company.
• Pay ~$1000 in fees to a company creation service (unless you do it yourself).
• State and show that you have a specific transfer in your bank account for a “capital amount”, normally $10,000.
• Wait >1 month for the creation process to happen (in total it took me 3 months).

This is for the Japanese LLC equivalent. For a Limited company I suspect it’s significantly more complex.

In contrast to this creating a company in the UK costs ~$20 can be performed entirely online and your company with be active in ~1 day. The US is similar, I used a lawyer that differed all fees and the process took <1 week.

On top of this, in Japan you will need to pay ~$1000 a year to keep the company active. Accounting fees also higher than elsewhere.

This is the most obvious thing that could be fixed. Make it cheaper, quicker and easier to create companies.

Seed funding

Summary: International seed funding is not available in Japan and there are essentially no domestic sources of seed funding. In contrast to this, in the US multiple sources of seed funding are available.

Richard’s article talks a lot about Angel funding. I recently received Angel funding for my startup. But I think individual Angel funding is probably only one of a number of drivers of startup creation in the US and at this point, may not be the most important. An argument can be made that this was more important in the creation of the Silicon Valley startup ecosystem. But even so, I doubt that tax breaks were a significant factor here.

Venture Capital

The US has a large number of seed funds (like YCombinator). I’ve interviewed with YC twice and have worked for 3 early stage YC startups, so I have a rough idea of how this works. I would guess that in the majority of cases YC is the first money into a startup. YC alone fund 400+ startups twice a year. All these companies are incorporated in the US. YC will fund companies physically based outside the US, but to my knowledge they have never funded a team which has stayed in Japan.

These kinds of accelerators/seed funds simply don’t exist in Japan. Those that do (e.g. DG Ventures) don’t invest in a large number of companies, and have various other issues…

Deep tech (science based) startups in Japan often work slightly differently. In the US these would often go the VC route or get government grants (see below). In Japan. however, VCs often seem go out and find some interesting university research then try and pair them with a non-technical CEO. The idea, I guess, is that the work is basically done and the CEO just needs to build a team and execute on this core IP… In the US deep tech startups generally start with a technical CEO… or at least one with a few successful exits.

Government Grants

The US has a government grant program called the SBIR, branded as “America’s seed fund”. Grants are often “themed” for example accepting applications in Genomics or COVID19 diagnostics etc. They work well for deep tech startups needing to perform proof-of-concept experiments. They can provide funding of $200,000 to ~$2,000,000. Funding is 100%, and for smaller grants the reporting requirements are minimal. It’s possible to start a company, get SBIRs and complete all your seed stage work without losing any equity in the US. Many US university spinouts have followed this route.

I’ve seen no such funding in Japan. There are NEDO government grants. These are I believe 50% matched funding. Once you get the grant you may be able to reclaim 50% of the cost of purchases from NEDO. From what I can tell each item requires competitive quoting. The reporting requirements are huge, and not covered by the grant. I would most likely reject a NEDO grant if offered one, the overhead is too high.

Angel Funding

There’s a much higher density of Angel funding in the US. There are multiple Angel networks, in some sense this is a function of a successful startup scene. Successful founders often want to continue to be involved in startup creation through Angel funding. I’m not aware on any significant Angel networks in Japan.

Miscellany

Summary: The US corporate environment is more mature. There are easier routes to getting patents filed and in general institutions are working to help new startups grow.

When starting my company in the US, I was able to find both corporate lawyers and patent lawyers who would differ payment until I raised $1,000,000. This makes it much easier to get things moving. Experienced IP lawyers are more readily available in the US, and filing in the US first is preferable in many cases.

Conclusion

The US has built a very mature startup environment which is unlike any other in the world. That being said, Japan has some clear barriers to startup creation which could be removed. Japan also has a strong high tech work force, which is paid relatively poorly. $100,000 might be entry level pay in the US. In Japan this offer would likely double the salary of many PhD level scientists. While there isn’t the density of engineers that exists in the Bay Area, with equal financing, you can likely hire experienced staff relatively easily at seed stage.

Many people have talked about the Japanese being “risk averse” and wanting to work for large companies. I’ve not seen much evidence of this. Rather, it seems that the required funding sources simply don’t exist in Japan, which limits startup creation.

I have a DR4000U photospectrometer that I bought as junk on eBay. After installing a new UV lamp the instrument powered up and passed power on self tests. I could take measurements and at a first glance things seem to be fine. However… on closer inspection almost all samples showed negative absorbance. The only way I could see this happening was if my samples were strongly fluorescent or they were somehow acting as an optical element reflecting light into the reference photodiode or perhaps focusing more light on the sample photodiode. But this proved not to be the case…

Using the instrument noise checks I also noticed that blocking the light path completely also resulted in negative absorbance. This makes no sense. Eventually using the noise check to look at the photodiode current I figured out something wrong with the instrument. The reference photodiode seemed to increase when I shined a light on the sample photodiode.

After comparing the wiring to another DR4000 I realized that someone had swapped the sample and reference photodiode inputs. This is relatively easy to do as they are both just connected using identical coax connectors. But what’s more interesting is that this /kind of/ works. That is all the power on self tests, which include scanning the grating and wavelength calibration (against the filter wheel references) worked.

So, this seemed like an easy fix. But when I switched the cables round the unit failed wavelength calibration. I then switched out the filter wheel with a working (visible only) DR4000 and the unit worked correctly. So, this instrument had a dead filter wheel, but somehow switching the reference and sample photodiodes tricked it into thinking more light was being detected than actually was, and the instrument passed calibration.

I will probably run this unit with the new filter wheel. But I’d also like to be able to replace the filters in the other unit. Below I’ve taken spectra of the good and bad filters. It might be possible to figure out what filters I need to get to replace these. Hopefully I can also dig up some further documentation somewhere.

Filters

I’ve numbered filters as shown in the image below (hopefully you can just about make out the scratches). This image is from a bad filter wheel. Physically the old but working filter wheel I have looks about as bad as this however, so I’m not sure you can tell an “ok” filter wheel from a bad one visually. In all likelihood the filters I have are all in poor condition however.

“Bad” spectra were taken from the instrument mentioned above. It failed wavelength calibration with “24: Monochromator error”. As mentioned, when the filter wheel was swapped with another unit the previously broken instrument started working. Each filter was illuminated with a Xenon light source and jammed up against a spectrometer head. Possibly not very accurate but may give some indication as to the nature of these filters.

The images below were taken from the “bad” filter set. I tried sticking them in the spectrophotometer and measuring their absorbance. The filters were however in the wrong orientation. Could possibly be instructive however:

I picked up an old USB optical power meter on eBay. This meter is on longer supported by Newport, and they don’t list it on their main site. If you go to their download site (https://download.newport.com) you can find software for the 841-P-USB, but as far as I can tell this is for a newer version of the instrument. The device does not appear to work with PMManager or any of the other software Newport currently support.

It looks like there were two versions of the 841-P-USB, and earlier “black case” version and a newer silver case version. The silver case version seems to have an expanded command set, and possibly runs at a faster baud rate.

However, you can talk to the 841-P-USB over serial and get data out of it. You should use 57600 baud (8N1). The 841-P-USB manual (for the silver version, which you can find on the site above) lists several commands. I used the following to take measurements:

*SOU	Zero offset
*CVU	Take single measurement
*CAU	Print continuous measurements
*CSU	Stop continuous measurements
*F01	Show device and sensor information
*PWC00000	Set the wavelength used for connection (e.g. *PWC00532)

If I do find any software that works with this instrument, or further information about the device. I’ll update this blog post.

This post was previously published on the substack.

The Lucira COVID-19 test is an “at-home” LAMP qPCR-like test. Essentially you receive disposable molecular test device, with embedded reagents for $55 which you use once and then throw away.

I was curious to better understand the instrument and went hunting for teardown pictures and patents. Luckily Brad Ackerman on pulled one apart and posted pictures on twitter. Some of these are included here with permission from Brad. The teardown pictures also closely match one of Lucira’s patents which gives us another source of information. The patent has some wonderful figures, and gives us a really nice exploded view of the device:

This really gives us a great overview of the instrument. We can see a single main PCB with contains LEDs, a photosensor, heating element, temperature sensor and microcontroller. The heating element is a simple PCB trace used for resistive heating. This is required get the reaction to ~60C as required for LAMP. Critically, for LAMP we don’t need do thermo-cycle, so cooling is not required. 

The photosensor sits in the center of the heating element, surrounded by LEDs. On top of this there’s a reasonably complex optical system:

The sample will flow down from the prep tube and be exposed to a reagents ending up in a number of fluidic chambers. This is kind of neat, because as we can see in the image above each LED is paired with a chamber. The light from the LED is reflected through the chamber and down on to a single photosensor.

This allows you to multiplex to some degree, using a single sensor and multiple LEDs which you can switch on/off. Naively you might think that for COVID-19 you only need a single measurement. But the FDA authorization makes it clear that the device also has a “Positive Internal Control (PIC) and Lysis Internal Control (LIC)” which must also show the expected results.

Looking at Brad’s teardown images, we can see that Lucira have only placed 4 of the possible 8 LEDs in the COVID-19 instrument. So Lucira could potentially expand the capability of the device to other targets/variants:

I wasn’t able to identify the photosensor used here, but I suspect it doesn’t cost more than $1, similarly the LEDs likely cost almost nothing. The microcontroller is an STM32F030C6, a simple 32bit ARM microcontroller with 32K of flash, 4K of RAM. This also usually costs ~$1. Overall I suspect the BOM cost is somewhere in the region of $5. The photosensor could potentially be the most expensive part if this turns out to be a high sensitivity photodiode…

The optical system appears to be all plastic, here you can see the light pipes. They show up as different colors, it’s possible there are embedded filters. You can also see the heat conductive paste that is used to couple the heating element to the rest of the unit:

The result of all this is a platform, where much like qPCR, you can monitor the amplification process in order to determine the presence/absence of your target. The patent shows some example experimental data:

While I suspect the electronics doesn’t cost much, I’m still impressed they can sell this for $55, as it’s a reasonably complex optical/reagent system. I’d be curious to know what their margins are. Overall I find the approach pretty neat. No doubt BARDA (who gave them $21.9M in 2018) must consider this a prescient funding decision.

Lucira IPO’d in February, but it seems that their share price has been rather unstable. It will be interesting to see what happens with Lucira and if they and gain adoption outside of COVID19 diagnostics.