[Week 3~4] DNA memory and the scaling of information storage

[Jie Xiang]

Folks, you've probably heard this week of the new Nature article online detailing the successful synthesis and readout with 100% fidelity of nearly 1MBytes of information on DNA strings. There is no new science involved as most of the processes are done using commercially available equipments. 

This paper (Linked at: http://www.nature.com/nature/journal/vaop/ncurrent/full/nature11875.html) presents a good argument on projected cost advantages of DNA storage, based on the Moore's Law-like downtrend of cost of synthesizing DNA base pairs (http://www.synthesis.cc/2011/06/new-cost-curves.html). 

Each DNA base pair is less than 40 atoms. I want to recall Richard Feynman's "Plenty of Room at the Bottom" speech: 

"Let us represent a dot by a small spot of one metal, the next dash by an adjacent spot of another metal, and so on. Suppose, to be conservative, that a bit of information is going to require a little cube of atoms 5 x 5 x 5 – that is 125 atoms....For each bit I allow 100 atoms. And it turns out that all of the information that man has carefully accumulated in all the books in the world can be written in this form in a cube of material one two-hundredth of an inch wide – which is the barest piece of dust that can be made out by the human eye. So there is plenty of room at the bottom! Don't tell me about microfilm!"

I'd like to hear your comments, maybe along these lines:

1. Memory density: you can buy a micro-SD card for $60 that stores 64GB. Would the DNA match this level? 

2. Energy required to readout DNA sequence - how does it compare to, say, reading out an SRAM or DRAM bit?

3. Access speed: this is somewhat related to energy. What are the limits? 

4. Note that living mechanisms have no problem reading/transcribing DNA information and even make copies on little energy. So the issue here is "readout" as in "read out aloud" as in macroscopically displayable.

5. Fellow classmate Sergio has posted last week about nano-magnetism: atom by atom, which can be compared here as another extremely scaled computer memory. How do these questions fare on that case? 

[chandra.christensen]

So first of all, I'm at home (and am having issues with the proxy), so I haven't read more than the first page of that particular article. But I did some searching on DNA based storage. 

According to this article, a harvard group was able to put Terabytes of data into a single gram. That's hugely, hugely, hugely more memory density than is commercially available. It's research, so obviously the cost isn't comparable. 

The above article makes a point that seems quite logical - due to the complicated nature of the reading and writing procedure, it seems unlikely that the read/write speeds / energy cost could ever even come close to MRAM. So this DNA based memory would be more useful for situations where you wanted to store a huge amount of data and access it relatively infrequently.

Perhaps in a very Avatar-esque science fiction future, all of our technology would mimic biology much more closely. In this case, I can see DNA based memory schemes being much more practical for instant read / write. (IE computers are based on E&M, so there's more of a one to one with our current memory. More biological computers could have the same advantages with DNA based memory?)

Spin based memory devices are a whole different story -- they are intrinsically very compatible with our current computing paradigm, and provide a very immediate potential to reduce costs (both space / energy / dollar) associated with storage and retrieval of information. 

(PS, thanks for posting those articles. I'd love to work on that kind of stuff once I'm a big kid in Grad School :) )

[matthias.engh]

I get concerned when reading in the Harward 700TB article:
'To aid with sequencing, each strand of DNA has a 19-bit address block at the start (the red bits in the image below) — so a whole vat of DNA can be sequenced out of order, and then sorted into usable data using the addresses.'
Since for practical applications some sort of memory architecture is desirable.

Also the mention of a hour long timeframe for reading a three billion(not tera) base pairs could turn out to be a problem.

Now to not be all negative, following the link 'eyeing up DNA as a potential storage medium' on the website, there is an article to be found, which mentions that DNA readily bonds with metal ions, which could be used for sorting the data. This article talks only about 'write once read many' type of memory, so maybe rather used for archiving data.

This article also states the read voltage to be only 2.6V and thus comparable to existing memory. Now this doesn't tell us the power or read out time, but unless they are excessive the energy is comparable too.

I think that how to store data in dna might not be the only problem. No one doubts that dna is small, but if it needs large read instruments/storage units or it's floating randomly arranged in a tank I don't think it will be in your next handheld device.

Cheers,
Matthias

[symay]

As science and engineering keep evolving, the paradigm here could be: why are we doing this kind of research at this point, when the FDA is going to ban it from getting to everyone? Millions of dollars have been spent for an in-vivo glucose sensor for example and theres not a single one that has an FDA approval. how many years ago did robotic implantable prothesis were commercially available? Maybe never? But the technology is already developed, right? As I am taking ECE247B (Bioelectronics) with professor Heller, as he mentions in his class: "there's not a single man-made procedure commercially available right now that can do a 100% accurate DNA sequencing". As this fact is a reality right now, I think that the development of DNA storage technology would benefit us a lot, but it may just never overcome the "hype cycle" showed on the first lecture. Writing and reading and sequencing DNA takes a long time and it is not 100% accurate. Apart from taking a lot of time, it is extremely expensive. From a recent study (I'll upload the paper link later) it is know that sequencing a single individual genome costs around $250.000 dollars (commercial equipment) and takes up to 5 days to finish with an accuracy of 96%. We as engineers may think, well, when i was doing my undergrad i approximated everything and 4% is not a lot, but let me tell you that it is a huge amount of data. even if its just 0.1% it will still be a huge amount of data lost. now just think about the problems of having DNA as a storage device at a large scale inside computers, gadgets and even inside us? It is impressive and at the same time a creepy feeling. We would be carrying information inside our bodies.would you be able to trust the Red Cross blood donation? How would encrypting the information in this context work?

[ajkerr]

I was thinking the same thing, Matthias, when I read the article.  An hour read time is insanely long.  It sounded like they were reading a lot of different things and then sorting them out with address tags.  That seems inefficient to me.  Also, the thing I was wondering about is the lifetime of something like this.  DNA is an organic molecule which requires a specific bonding structure to retain the useful information.  As we all know, DNA is susceptible to damage and degradation by environmental factors such as temperature, pH, radiation, etc.   I feel like a lot of care would have to be taken not to accidentally damage the DNA molecules.  Care that is not required when dealing with more robust silicon transistor circuits.  I mean, I once washed an SD card in the washing machine accidentally and it still works. 

This is a really neat idea, though.  I don't think we'll be seeing a lot of commercial products with this technology any time soon, especially considering the cheap cost/GB of flash cards, magnetic memory, SSD's, and even ultra-fast forms of quick-access memory like DDRAM for processors.  But maybe someday in the future.  They have to start somewhere, right?

-Amanda

[infinitexh]

Hi all, after seeing all of your interesting points on DNA memory I just couldn't resist but jump in; however just a disclaimer that I am by no means a biologist (the last bio class I took was two years ago), so please excuse any inaccuracies in my posts if any ;)

Personally I'd love to see more incorporation of DNA usage in memory storage; while most of the pro and cons have been addressed in the previous posts, I feel that we have neglected something also very far fetched but ultimately interesting uses for DNA memory storage: directly integrating the DNA decoded information into other biological mediums, even our brains, possibly leading to a matrix-esque far off future. 

DNA decoding is a relatively understood process with the key involvement of certain restriction enzymes which process DNA strands in vivo  (for those who are more interested =) http://seqcore.brcf.med.umich.edu/doc/educ/dnapr/sequencing.html) And since this process is all done in vivo, why can't we do it in our own body and somehow transfter it into our own brain? Temporary storage of DNA in our tissue (skin for example) has already happened (as mentioned in your guys previous articles) and really the only key challenge that I can see as a nonbiologist is the translation and incorporation of the decoded message directly into our long term memory unit in the brain. Versus our previous inorganic systems for storage, using DNA as a storage unit opens up many other fundamental doors that inorganic materials simply cannot give us - and exploring this new dimension of utility i feel, should be of utmost importance when the fundamental paradigm of the device itself has changed so drastically.

While my suggestion is certainly extremely far fetched and I know no neuroscience friend to ask them about even the probability of my idea. This idea does seems however, to be just as "far fetched" as other fields that are already being researched for the far future but with certain deadly stumbling blocks like topological insulators, and various quantum spin hall effect projects. And ultimately I think there are endless merits in investigating extremely fringe sciences (or sciences that even grad students like us consider "fringe" =) ), notwithstanding with our central philosophy as scientists to understand the world around us.

Finally, even if we only aim at replacing our current magnetic based memory technology with DNAs, there are also, in addition to previously mentioned advantages, other advantages present such as reliability and easiness in storage (as mentioned in Amy's article) and the ability to bio degrade. E-waste has been a huge problem in 3rd world countries and costs tremendous amounts of money to take care of in 1st world countries, our current movement towards bio-devices (OLEDs for example) means biodegradation in the future and at the very very least prolong our longevity on this planet. And given DNA's extremely dense memory capacity, I feel appropriate that it is the perfect archival unit for some kind of "human knowledge" library. Not to be pessimistic, but humanity now more than ever does possess the ability to destroy itself utterly and completely; and having an archive that contains the entire collected knowledge of the human race is but another good investment/insurance that we can make to ensure the survival of our own specie into the far future as well.

So there it is, dreaming like a mad scientist; is this what we all do??

Best,

Peter

[matthias.engh]

Without excluding the possibility of DNA based memory, it seems as if DNA memory's primary use could be within bio-micro-robots as the following two articles point out

- http://www.sciencenews.org/view/generic/id/340900/description/DNA_used_as_rewritable_data_storage_in_cells

- http://www.sciencenews.org/view/feature/id/347263/description/Factory_of_Life
, such as bio assasin drones.

To try and not derail to far from the initial post, the compatability of DNA with the bio organism removes the need for silicon to organic 'translators' and thus in a sense the size of the memory in this case. Additionally, without knowing too much about biology, I assume the energy required to read the data can be freed using 'conventional' bio mechanisms. This of course is not directly an answer to the questions in the initial post and not at all about 'reading out aloud'.

In the first article, in the comment section there is an interesting reply to Roger Brents initial statement on replacement of silicon through DNA memory though, and I quote:

'It is an interesting idea, but it seems like an impractical approach. Keeping track of eight bytes of information will require 8 different DNA sequences plus the DNA, RNA, and protein for 8 different recombinases, each one tracking a different DNA sequence.

Although I cannot suggest an improvement at this point, it seems like a fundamentally different approach is needed.

As this is outside of my knowledge spectrum I have no opinion on it's correctness however.

Cheers,
Matthias

[shichuan1989]

Hi All:

I think it is possible to achieve the Memory density. First, At theoretical maximum, one gram of single stranded genetic code can encode 455 exabytes of information. That's almost half a billion terabytes, or 4.9 * 1011 GB. (As a point of reference, the latest iPad tops out at 64 GB of storage space.) DNA strands also likes to fold over on top of themselves, meaning that, unlike most other digital storage media, data needn't be restricted to two dimensions; and being able to store data in three-space translates to more free-space.

 

Besides I have read a paper Written by Kosuri and his team:

http://www.sciencemag.org/content/early/2012/08/15/science.1226355.abstract

To demonstrate the vast potential of DNA storage, Kosuri and his team used just shy of 55,000 159-nucleotide chunks of single stranded genetic code to encode a 5.27-megabit book, containing 53,426 words, 11 jpg images and one JavaScript program. They then proceeded to use next-generation DNA sequencing techniques to read it back. (For those who need refreshing, nucleotides are the individual building blocks that, when joined together, form strands of DNA.)
 
5.27-megabits probably doesn't strike you as a lot (that comes out to roughly 660 kilobytes of information, about what you'd find on a 3.5" floppy from the 80s), but it's impressive for at least three reasons:
 
One: It positively crushes the previous DNA-storage record of 7,920 bits.
 
Two: The novel encoding method employed by Kosuri and his colleagues allowed them to address issues of cost and accuracy, two long-standing technical hurdles facing DNA storage:
 

The major reason why this would have been difficult in the past is that it is really difficult to construct a large stretch of DNA with exact sequence, and make it cheaply.  We took an approach that allows us to use short stretches of DNA (basically by having an address (19 bits) and data block (96 bits), so each short stretch can be stitched together later after sequencing. Using short stretches allowed us to leverage both next-generation synthesis [for writing data]… and next-generation sequencing [for reading data] technologies to really lower cost and ease.
 
Three: It offers a compelling proof of concept that DNA can be used to store digital information at remarkable densities. "What we published in terms of scale is… obviously small compared to commercial technologies now," explains Kosuri, but "using our method, a petabyte of data [one petabyte = 1,024 terabytes] would require about 1.5 mg of DNA." Since that genetic information can be packaged in three dimensions, that translates to a storage volume of about one cubic millimeter.

[samontoy]

DNA Memory Storage

I read this paper quite a while ago, and although its a cool project... the present state does not make it a promising technology which industry will likely begin developing as a product any time soon. 

Unless I misunderstood the data in the plots, the cost for 1Mb is 125 to 500 dlls. If we quantify this to existing memory technologies, say a 64 Gb SD Micro Card which costs about 64 dlls then a similar technology based on present DNA memory storage would cost about 125000 to 500000 dlls for 1 Gb; therefore, someone would have to pay a grand total of 32 million for a 64 Gb storage device which could only be accessed a very few times during its lifetime. I’m well aware that this technology is novel and has not been optimized as a consumer product. Industry rarely begins adopting new methods for memory storage if the present technologies are performing very well. 

Let us examine a present case where this is likely. I’m quite sure everyone has heard of STT-MRAM and are aware that there are many companies attempting to build this memory to replace the standard RAM technologies. This change is not occurring because of problems being encountered in scalability of each bit (as is being lived in hard drive technologies - where industry is facing problems as storing more data in magnetic grain particles). Both technologies can be scaled quite a bit more before we begin to encounter thermal stability problems of the bits. It has nothing to do with the cost of making either MRAM or RAM devices (although in the long run the cost for MRAM will likely decrease).... Instead, it has to do with the fact that this technology is specifically being designed to be imbedded in mobile technologies. Over the past decade, we have all been part of this complete redefinition of what mobil technology is: from smart-phones to tables and everything in between. However, all these technologies use power and we all want these tools to last longer. The primary reasons why MRAM will replace RAM is because: (a) It does not require current to maintain the bits of information (unlike RAM), (b) If we scale a RAM bit the current does not scale, but scaling a MRAM bit means reducing the magnetic material and thus the switching field of each cell, i.e. the needed current to change a bit. Presently, tape memory is probably more expensive than any other type of conventional memory, it can not be written nor read very fast (~200 Mb/s), nor is the amount of tape required to store information really small. However, it is a very robust technology which is not limited by read-write access as DNA based memory. 

Another surprising fact about magnetic based memory is that this technology in overall is fairly inexpensive and has always been so. Last time I checked a conventional hard drive costs a tenth of what a consumer usually pays, yet the cost of the manufacturing process machines is in the billion dollar range. Most companies that go into this business don’t make a profit for a few years, and I know that companies do not like to invest in modifying the manufacturing process because it costs money. From conversations with people from the hard drive industry and my own experience, I know research and development is bounded by existing methods of manufacturing. This is why HAMR (Heat Assisted Magnetic Recording) is favored in industry: It does not require redesigning the manufacturing process of the media, but only of the read head. If your interested here’s a link that can give you some information on HAMR: http://nanomag.ucsd.edu/wp-content/uploads/2011/12/Kryder-review1.pdf . The point that I’m attempting to get across is that there’s little possibility for industry to invest the resources to reduce the cost of DNA memory by several factors to make it a viable memory storage option. It’s more likely that self assembled media or patterned media will be favored because they could be incorporated into HAMR. 

Earlier this year, atom by atom memory storage was introduced at the Intermag/MMM Conference in Chicago. By the way, this project was an effort of development of nearly 20 years. See previous post of ‘atom by atom’ for the paper. I can’t compare the cost between this technology and DNA based memory because I don’t know the what the cost is estimated for a single bit. However, we can estimate the amount of memory that could be stored: The point of this technology is to store a single bit of memory in a ferromagnet atom. Let us suppose that instead we stored it in a magnetic grain (many many atoms)... At the present moment we currently store 1 bit in about 40-60 grains in PMA media, this means that you could expect to see a 40 to 60 times more memory storage in you typical hard drive (a 2Tb HD could become 80 to 120 Tb HD). This without any doubt is a lot of space !  Probably the main problem I can foresee that will be required to push this technology forward, is the development of tools to move atoms efficiently at a fraction of the cost it currently costs. We all saw in class that the high tech e beam writer at Nano3 costed about 2 million dollars and we can all conclude it’s not a robust system. So there’s no doubt we are a few decades away before we see a new and less expensive methods being developed that can perform the same task at a fraction of the cost. Personally, I found this technology to be awesome !  I hope to see this type of memory storage become a consumer product in my lifetime. 

In overall, I believe that DNA based memory storage is novel and I think it’s truly incredible that we can manipulate a string of DNA to encode information. I do not have a strong chemical nor biology background, so I find it difficult to judge this technology appropriately. That is to say, I don’t know from personal experience how you go about making a DNA string and how you read information. Encoding information is not a problem because mechanism can be devised by new logic structures. This is the reason I compare it to magnetic storage which I’m familiar with. In case someone is interested in another form of memory storage, there one that was introduced a few years ago: Domain Wall race track memory by Stuart Parkin:http://www.sciencemag.org/content/320/5873/190.full . Also a cool idea... 

Cheers, 

Sergio