QuantumSi

There’s very little public information about QuantumSi. Sequencing isn’t mentioned on the company website [1]. QuantumSi is part of the 4Catalyzer group of companies, founded by Johnathan Rothberg. Rothberg of course, previously founded sequencing companies 454 Life Sciences, and Ion Torrent. QuantumSi’s patents mostly list Rothberg as an inventor and are largely DNA sequencing related.

Technology

The QuantumSi approach as presented in patents appears to be similar to the Pacific Biosciences Zero Mode Waveguide platform:

QuantumSi and PacBio approaches from [2] and [3].

In this approach a polymerase is tethered to the bottom of a nanoscale well (ZMW). The well is illuminated, however because the well is smaller than the illumination wavelength the photons can not fully enter the well. Instead, you get an evanescent field confined within a few nanometers of the surface.

This is not unlike TIRF microscopy, which similarly confines illumination to within a few nanometers of a surface (and is used in Illumina sequencing [4]).

To sequence a strand of DNA the polymerase incorporates fluorescently labeled nucleotides. There are no terminators so nucleotides are incorporated in real time. During incorporation the fluoresces are cleaved and released. This results in a brief “flash” as the label is excited under the evanescent field.

These flashes need to be monitored in real time. This is were the QuantumSi and PacBio approaches appear to differ. Looking at RS and PacBio chips, it looks like PacBio moved from external camera in the RS to a fully integrated chip in the Sequel:

PacBio patents [5] suggest that they may use a fresnel lens integrated under the ZMW. Under the lens are multiple (appears to be 2) filters and sensor:

Integrating a fresnel lens onto a semiconductor device seems really neat. However I suspect this makes for a significantly non-standard (and more expensive) fabrication process.

The QuantumSi patents differ in terms of detection. Rather than using filters to detect labels with different excitation wavelength, the QuantumSi chips use intensity and decay rate to distinguish between different labels [6]:

I guess these may be standard fluorophores, but I’ve not dug into the patents to find out exactly what’s being used. While the patent mentions excitation “intensity” I’d assume that the photosensor responds with differing intensity to different wavelengths, so it could just be this that results in differing intensity between the fluorophores. The rate of fluorescent delay, provides a secondary signal to help distinguish between label types.

This approach most likely results in a cheaper chip, that can be more easily manufactured using a standard manufacturing process (as no lens or filters are required).

Given Illumina’s recent acquisition of Pacific Biosciences the potential for a low cost competitor to PacBio using a similar approach is interesting. I’ll be keeping an eye on QuantumSi, Rothberg has a track record of executing well, and I suspect we will hear more in the not too distant future.

Notes

[1] http://quantum-si.com/

[2] https://patents.google.com/patent/US20170362651A1

[3] https://patents.google.com/patent/US7973146B2

[4] https://blogs.swarthmore.edu/Illumina+GAIIx+Teardown/?p=27

[5] http://www.freepatentsonline.com/20180180548.pdf

[6] http://www.freepatentsonline.com/y2017/0349944.html

2018 DNA Sequencing Raises and Acquisitions

As we’re heading toward the end of 2018, I thought it might be interesting to put together a list of DNA sequencing companies that have raised in 2018. I’m aware of 14 companies that either raised or were acquired this year. Excluding acquisitions this totaled 368MUSD of investment.

Company Amount (MUSD) Further Info
Centrillion Biosciences 27.5 info
Direct Genomics 34 info
Genapsys 32.5 info
iNanoBio 3.3 info
LaserGen (acquired) 105 info
Omniome 60 info
Oxford Nanopore 140 info
Pacific Biosciences (acquired) 1200 info
Quantapore 15.55 info
Quantum Biosystems 8.8 info
QuantumDX 12 info
Stratos Genomics 20 info
Two Pore Guys 2.9+4.6? info
Universal Sequencing Technology 8 info

A couple of other companies raised toward the end of 2017:
Cygnus Biosciences, 19.6M USD info and
Base4 Innovation Ltd, 6.5M USD, info.

Thoughts on Ion Torrent and the iSeq

Recently I’ve been thinking about the economics of semiconductor sequencing, in particular approaches like Ion torrents, and the iSeq which use a relatively large die which gets thrown away with the rest of the flow cell.

In this post, I’m using to review aspects of Ion Torrents development, I don’t cover the technological approach (which is well described in their Nature paper [3]).

Business (to acquisition)

Ion torrent was founded in 2007 by Jonathan Rothberg (no doubt building on his prior success with 454). They raised approximately 60MUSD in funding (from Bay City Capital among others).

Ion torrent was acquired only three years after it was founded in 2010. Life Technologies initially paid 375MUSD with potential for another 350MUSD in milestone payments. Milestone payments were paid in 2012 [1], making the final acquisition price 725MUSD.

It’s surprising how soon acquisition came after funding, both rounds which happened less than a year before acquisition. Given the platforms reliance on semiconductor fabrication, this doesn’t seem like enough time to get much done (iterate over a new chip for example).

I therefore assume that prior to the rounds listed there was probably some seed funding, perhaps supplied by Rothberg himself.

I’d also guess that the investment rounds were at a valuation of 100 to 200MUSD. So the exit was at x4 the last round price. Given how quickly the exit came, this seems pretty good.

Business (current)

Ion torrents market share is something like 10%. Their niche is a relatively cheap benchtop instrument aimed at small labs. They’ve also developed a number of diagnostic applications around the device.

The Ion Proton uses a chip with a ~20mm square die. Based on my estimates this probably costs at least 200USD to fabricate. Given that runs cost 1000USD, I’d guess their margins are lower than Illumina’s, probably 50 to 60% on consumables? [2] (compared to 90% for Illumina).

The ion torrent 318 chip is smaller ~1cm^2. At minimum I’d guess this costs 50USD. The kit here sells for 625USD. It’s therefore possible margins are better here.

It’s also interesting to compare the 318 chip to the iSeq (illumina’s new instrument). They both have similar throughput. The runs both cost 625USD. The iSeq die looks smaller however, which may allow higher margins on the iSeq.

I’d also guess that the iSeq uses a standard CMOS image sensor, not developed exclusively for Illumina. This most likely reduced development costs, and means that Illumina can take advantage of a more robust supply chain, and economies of scale that come with a commodity part.

In short, it feels like Illumina maybe able to squeeze Ion torrent a little more with the iSeq, given their costs are probably lower.

Notes

[1] https://www.genomeweb.com/sequencing/life-tech-initiates-second-milestone-payment-ion-torrent#.W_B_SRaRUlQ

[2] There’s a report from system plus on the Ion 520 chip, it appears this contains detailed cost estimates, however it costs 3500Euros…

http://www.yole.fr/THERMO_FISHER_ION_TORRENT_TechnologyOverview.aspx

[3] https://www.nature.com/articles/nature10242

How much does it cost to ship 1cm^2 of silicon?

I’ve been trying to figure out how much (approximately) it costs to ship 1cm^2 of silicon from public data. Public costs for 300mm wafers seem to be largely unavailable. I’ve seen costs of 400USD for a unprocessed wafer, 500USD for a processed wafer under an older process, to several thousand dollars under leading edge processes. But no official pricing.

In order to get some kind of citable reference I decided looking at commodity ICs, of known area might give some indication of the costs associated with the production of a packaged IC. In particular I decided to look at image sensors where gross margins appear to be relatively low (20 to 30%) [1].

So, I extracted all image sensors with pricing from digikey [2]. I used the listed pixel size and array size to estimate the size of the die (actual die size is no doubt slightly larger than this, but hopefully this is good enough to give a reasonable estimate).

I then used the die size to calculate the cost per mm^2, and the cost per cm^2. The plot below shows a histogram of the calculated cost per cm^2:

As you can see there’s a peak around 40/50USD. The cheapest part, that has some stock works out at 52.12USD per cm^2.

The cheapest cost per cm^2 was 12.88USD [3], however this part is not in stock, and on 14 week lead. I suspect that the part is not actually available.

So, based on this rough estimation I’d guess that a cost per cm^2 of ~40USD isn’t completely ridiculous (given the relatively low margins for these low cost image sensors).

Notes

[1] https://www.reuters.com/article/us-omnivision/omnivision-reports-weak-profit-as-margins-shrink-idUSBRE84U1A220120531

[2] www.digikey.com (US). Random selection of awk used:

cat CA.csv | grep -v Obs | grep -v Discon | grep -v Last | grep -v Active | grep -v “,-,” | grep -v “,-” > CAS.csv

awk ‘BEGIN{FS=”,”;}{print $3 }’ CAS.csv | awk ‘{print $1 ” ” $3}’ | sed “s/[^0-9. ]//g” > pixsize

awk ‘BEGIN{FS=”,”;}{print $3 }’ CAS.csv | awk ‘{print $1 ” ” $4}’ | sed “s/[^0-9. ]//g” > arraysize

awk ‘BEGIN{FS=”,”;}{print $1 ” ” $2 ” ” }’ CAS.csv > tp

cat CAS.size | grep -v Reel > CAS.size_nr

awk ‘{print $1 ” ” $2 ” ” (($3/1000)*$5)*(($4/1000)*$6)}’ CAS.flat > CAS.size

awk ‘{print $1 ” ” $2 ” ” $2/$3 ” ” ($2/$3)*100}’ CAS.size_nr | sort -n -r -k 4 > CAS.calc

cat CAS.calc | awk ‘{print $4/10}’ | awk ‘BEGIN{FS=”.”}{print $1}’ | awk ‘{print $1*10}’ | uniq -c | awk ‘{print $2 ” ” $1}’ > hist.csv

Files: data

[3] OV07726-G04A-ND Unit cost 1.425USD for 2276 units. 6um pixels, 640 by 480 array.

[4] AR0330CM1C00SHAA0-DP1-ND 2304H x 1296V 2.2µm x 2.2µm, 7.85USD for 152 units.