selenite0

Chris Phoenix of the Center for Responsible Nanotechnology wrote a detailed description of a nanofactory using an array of molecular assemblers to produce macro-scale products.

I think the basic design can be made to work, but it's going to need a lot of support infrastructure, and it still won't be able to do the high-speed duplication and bootstrapping the paper describes. Phoenix refers to this capability to justify his opinion that a nanotechnology breakthrough could rapidly transform the world. I dispute that.

An informal write-up of my take on the design paper is on our website (with supporting calculations here). This post is open to comments on my paper or the original nanofactory paper. Non-LJ users, please use the anonymous commenting feature and sign your name or handle.

[Discussion in comments and in this post.]

Current Mood: thoughtful

Flat | Top-Level Comments Only

From: (Anonymous)

Karl's comment that some issues were only covered at a very high level is absolutely correct. The paper was already far too long (about 80 pages). It would be great if, as Karl suggests, others would help fill in the design.

But I would caution that a common mistake among those who haven't studied molecular manufacturing is to choose a difficult design option and then assume that this choice limits the design. For example, Karl is worried about whether carbon can be used to make electromagnets. But nanoscale electrostatic motors have a vastly higher power density (about 8-10 orders of magnitude) than today's electromagnetic motors. Once a nanofactory is buildable, then certainly a motor block made of many nanoscale electrostatic motors ganged together will also be buildable; and its mass will be negligible. There is no need for electromagnets.

Other points Karl missed in his analysis:

1) The nanofactory has a rather high power draw, ~200 kW. It will not care about external cooling, since heat transmission through the shell will be a tiny fraction of the internal use. (And it will not use a three-prong plug!) This is why I talked about solar collectors in section 7.5: once you start building thousands of nanofactories, you will need a new source of power.
http://www.jetpress.org/volume13/Nanofactory.htm#s7.5

2) Likewise, the fluid will not come from a reservoir, but recirculated through a rather large cooling plant. I recommend thermionic cooling, as that has been done with diamond electron emitters. And no, I have not calculated the mass of this cooling plant, but thermionics work best when they're quite thin.

3) The most delicate operations (the mechanosynthesis) will use the smallest equipment. Equipment at this scale has a vibrational frequency measured in GHz. External sonic vibrations will have no effect.

4) I did calculate data distribution along all the large levels, concluding that six rod-logic rods would be sufficient; see 8.1. Also, I calculated power distribution (8.2), though I see I didn't spell it out as well as I should have. Basically, a rotating diamond rod can transfer power at >20 W per square micron; it scales linearly with area, see http://nanomedicine.com/NMI/6.4.3.4.htm . So the 200 kW power load requires only a cross-section of 10,000 square microns, or 100 microns square. That's for the whole factory; a 1-nm rod would easily drive a production module.

5) There won't be stagnant regions of fluid adjacent to the fabricators, since flow there is laminar. I forgot to calculate it in the paper, but I calculated it later: more than sufficient feedstock can diffuse to keep all the fabricators supplied, even at low feedstock concentration. And the rate of flow of cooling fluid will circulate far more feedstock than the rate of fabrication requires. I explicitly noted that fluid might stagnate in gaps between modules (4.3). This is not hard to prevent; just bridge the gaps with thin walls to contain the fluid. The fluid will move fast enough that micron-scale blobs of higher density will not have nearly time to settle out.

Continued...

From:

selenite.livejournal.com

I was trying to come up with a minimum level of infrastructure. This meant tapping into the existing power grid and trying to minimize the thermal load. If you want to have a dedicated 200 kW power plant and a cooling unit that can keep it ice cold under any ambient conditions you're having a much more complicated and heavy set of support equipment than I'd envisioned. This is going to add to the amount of time to replicate the nanofactory, since a new factory is useless without the full infrastructure set. I think having that kind of independent set-up will significantly slow the autoreplication process. The time to replicate the nanofactory proper will probably turn out to be a small fraction of the total production timeline.

As for vibrations, while the assemblers may be too small to be affected any vibration will probably come close to the critical frequency for one of the assembly stages, so you still need to isolate it while assembling the product. The vibration isolator may be designed to take advantage of how the problem frequencies change as the product blocks proceed through assembly, so it only needs to block a portion of the spectrum at a time.

From: (Anonymous)

Other arguments:

1) Molecules being imported will be surrounded by the binding site, and flow will be far slower than thermal velocity, so macro-scale intuitions of them being knocked loose by flow simply don't apply. Likewise, binding of molecules in a binding site shouldn't be very dependent on pressure of a few atm.

2) My description of assembly as happening "slowly" referred to the last few fractions of a micron, to allow for temperature equilibration. Assembly operations may be slow, but transport can be faster. Finally, the assumption of only one gantry crane per tube is unnecessary unless one is looking for a crippled design; I was assuming a crane waiting by each inlet, and finished blocks being handed off from one crane to another, so that they would all move in parallel.

3) The step drives require only the equivalent of a couple of rod-logic gates to implement; rod logic is reversible by default assuming the designer knows how to use it. This is a minimal system impact. I think my efficiency assumption is quite safe.

4) "Routine mechanical malfunctions" requires either breaking of chemical bonds under low strain, which is vanishingly unlikely, or a failure of automation design allowing too much slop/unpredictability, which can be debugged and corrected.

5) Since the nanofactory is internally held in tension, unfolding does not require a lot of actuators, but merely inflation. Internal mechanisms, even the largest, could be held in place by Van der Waals forces until they were powered up and pulled themselves loose. Thermal insulation isn't really necessary, but if it were, two square-meter panels can stacked with vacuum between, for quite good insulation.

6) Karl claims that nanofactory development cannot be started until the first fabricator is completed. But we will have good simulations of the fabricator's mechanical properties, with bounded uncertainties, and the nanofactory design doesn't have to provide exactly the right amount of vibration or coolant flow--it can be overdesigned a bit.

7) MNT advocates are aware that the manufacture of simpler devices will be significant by today's standards. However, the last 1% of engineering closure of an automated autoproductive system drops the cost-per-feature by at least three orders of magnitude, dwarfing all previous rates of change and technological capabilities. I certainly hope we are inspired to get ready for full MNT by the earlier products. But I think the effect will be the opposite: we will think that change can't happen any faster than it's currently happening. And even a year is *not* enough time to make policy.

Chris Phoenix, CRN

From:

selenite.livejournal.com

I was assuming a crane waiting by each inlet, and finished blocks being handed off from one crane to another, so that they would all move in parallel.

That would bring the assembly time down to 23 minutes, still much more than a "few seconds". An isometric showing the last 2-3 stages with the cranes in action would make that a lot clearer.

4) "Routine mechanical malfunctions" requires either breaking of chemical bonds under low strain, which is vanishingly unlikely, or a failure of automation design allowing too much slop/unpredictability, which can be debugged and corrected.

You can't prevent all errors from occurring--the universe is far too good at finding ways to sneak them in. In the nanofactory it's most likely to suffer from versions of foreign object damage. Contaminants in the feedstock, reactive molecules leaking into the vacuum areas, and fragments broken off experimental products will all lead to problems. Making a system reliable requires making it able to handle errors once they occur, not trying to prevent them all.

Since the nanofactory is internally held in tension, unfolding does not require a lot of actuators, but merely inflation.

I still can't understand how this would work. If the structure is built in tension to automatically unfold when released, you still have overhead in the latches to hold it in place initially and springs for all structures that won't automatically unfold. If inflation refers to using gas to expand it I'm even more confused--a collection of rigid planes can't stretch like a balloon, so keeping the gas sealed in would take even more overhead structure.

we will have good simulations of the fabricator's mechanical properties, with bounded uncertainties

Simulations can't identify "unknown unknowns", which are always the biggest problems in new technology developments. A simulation of a brand-new technology will always have gaps until there's real data to work with. Assemblers won't be an exception to this.

the last 1% of engineering closure of an automated autoproductive system drops the cost-per-feature by at least three orders of magnitude, dwarfing all previous rates of change and technological capabilities

Closing the first autoproductive system will give you something that can reproduce itself given a long timeline, and whose products will not be at all competitive with dedicated factories producing more complex products than the nanofactory can put out (ie, not requiring unfolding, able to use whatever mix of elements is optimum for the task, taking advantage of the learning curve from previous developments). So it will be an existence proof, not a transformative device. Once you get a little way along the learning curve you'll have a useful autoproducer, but that's going to take time. It won't be a "Big Step."