Philip Turner @philipturnerar Twitter profile

Pinned Tweet

Philip Turner

2 months

Nanomechanical Computing: When Matter Becomes Information A research project on the smallest computers permitted by the laws of physics. Revealed new ways to design and simulate nanoscale machine systems. Minecraft, VLSI, and more. #MNT #APM #nano

2

13

82

Last Seen Profiles

@LClaw85

@leisurely_w0lf

@YukinoYi43660

@billyboypromise

@carrot_fi

@rahaf32744118

@arconnectio

@pengen_stw

@cibeerrrnetica

@PropheteMalumba

@pengen_stw

@kswavyy

@zhangxufan9999

@minigirlmizuoto

@AsadKha5107867

@9ali_

@honopan_yume

@Twonecannady

@hangover_pixie2

@jjohnsonj1

@KTTA_00

@ezgipeksen

@twicrll

@OneDex_X

@i_ambahar

@ParkerSien80948

@Yoonnababy

@Margaretamelia0

@mansoor_nisar6

@rennslyaer

@ABRLive

@RomanBJ_

@rbressane

@tfquila

@kavipottachi

@bllingham

Philip Turner

@philipturnerar

8 months

Atomically precise NOR gate, welcome to the world of computing. Any combinational logic circuit can be built with you. [1/n] #MNT #APM #nano

Philip Turner

@philipturnerar

9 months

Ok Eric, challenge accepted. Mechanical transistor 215x more compact than the TSMC N3 node.

10

18

274

53

238

2K

Philip Turner

@philipturnerar

7 months

21,264,880 atoms and 2,528 parts. This is nanofactory material. [1/19] #MNT #APM #nano

Philip Turner

@philipturnerar

7 months

How would an atomically precise computer be manufactured? #MNT #APM #nano

7

12

136

26

92

619

Philip Turner

@philipturnerar

1 year

A preview of what's to come: fundamental algorithmic improvements to FlashAttention, something not even done by OpenAI Triton.

9

27

274

Philip Turner

@philipturnerar

9 months

Ok Eric, challenge accepted. Mechanical transistor 215x more compact than the TSMC N3 node.

10

18

274

Philip Turner

@philipturnerar

1 year

I reverse-engineered the M1 GPU, now I reverse-engineer GPT-4. If GPT-4 is 8 x 220B as @realGeorgeHotz says, it would shard across one Nvidia SuperPOD while consuming 440 GB / 640 GB memory. It would inference in >27.5 ms, meaning <36 tokens/s. Could someone measure the [1/n]

5

12

266

Philip Turner

@philipturnerar

11 months

Superlubricity: sliding with near-zero friction, the mechanical analogue of superconductivity. 37k atoms, 1 ns. Nanomachine designed in collaboration with @mechadense . [1/11] #mnt #nanotechnology #apm

11

25

227

Philip Turner

@philipturnerar

3 months

Final assembly step for a Ge-C flywheel system composed of two parts 91,533 atoms, 40 picoseconds, 298 K

8

15

213

Philip Turner

@philipturnerar

4 months

Theoretical upper bound for logic density of mechanical computers: 2,300 atoms per switch, or ~177,000 MTr/mm^2.

12

13

180

Philip Turner

@philipturnerar

7 months

How would an atomically precise computer be manufactured? #MNT #APM #nano

7

12

136

Philip Turner

@philipturnerar

1 year

It took months on both the theoretical and practical sides, but the effort put into has paid off. A new rendering algorithm optimized specifically for atomically precise machines. #nanotechnology #gpgpu

GitHub - philipturner/molecular-renderer: CAD program for molecular nanotechnology

CAD program for molecular nanotechnology. Contribute to philipturner/molecular-renderer development by creating an account on GitHub.

github.com

1

14

110

Philip Turner

@philipturnerar

1 year

It's official. Now I'm up there with xformers and Triton.

3

11

106

Philip Turner

@philipturnerar

8 months

@LukeFarritor I have high-level plans for a full microprocessor, which I can simulate in full atomic detail. Not the entire thing, just each subsystem in isolation (a few million atoms each). Got access to some very powerful compute clusters.

2

0

82

Philip Turner

@philipturnerar

6 months

Nanomechanical robot arm with two grippers. #MNT #APM #nano

4

8

82

Philip Turner

@philipturnerar

7 months

Tripod tips on Au(111), for performing atomically precise manufacturing onto a scanning probe tip.

6

8

71

Philip Turner

@philipturnerar

7 months

This is a manufacturable strained shell structure. Using selective reconstruction of different (100) crystal planes, you can encourage faces to expand or contract. Without requiring non-carbon atoms like nitrogen and sulfur. Only 5- and 6-member carbon rings. #MNT #APM #nano

2

6

70

Philip Turner

@philipturnerar

4 months

Mechanosynthesis of a carbon atom onto a graphene build plate

8

3

71

Philip Turner

@philipturnerar

7 months

What can this build? A very dense array of mechanical logic. This array uses the same volume and energy as one “3 nm” transistor. Moore’s Law is nearly over for transistors, but classical computers remain important. Nanofactories are the only way to continue Moore’s law. [19/19]

8

3

65

Philip Turner

@philipturnerar

7 months

This was an exploratory engineering project on designing large nanosystems in atomic detail. It exposed new techniques required to construct and iterate at the million-atom scale. [3/19]

1

64

Philip Turner

@philipturnerar

8 months

For a medical nanobot or a self-replicator (part of the technology that enables mole-scale manufacturing), an entire ALU must fit inside a million or two atoms. TSMC won't cut it. Instead, we need mechanical computing, which is more compact than electronics. [6/n]

2

3

60

Philip Turner

@philipturnerar

8 months

This is quite manufacturable, using 2.5D lattice-aligned parts instead of strained gears and bearings. Imagine an AFM scanning over the surface like a macroscale 3D printer. Build each part, then use van der Waals forces to auto-align them. [4/n]

3

0

58

Philip Turner

@philipturnerar

1 year

I’m rewriting my FlashAttention code base from scratch, and I will get it right this time. Open-source from day one →

3

6

59

Philip Turner

@philipturnerar

1 year

First open-source implementation of FlashAttention on an architecture besides Nvidia, and on hardware accessible to the average person. 💥

Create old-flash-attention.metal · philipturner/metal-flash-attention@8de79fe

github.com

1

3

57

Philip Turner

@philipturnerar

8 months

Functional completeness - both NOR and NAND gates share this property. [2/n]

Functional completeness - Wikipedia

en.wikipedia.org

1

0

57

Philip Turner

@philipturnerar

1 year

276.1% speedup

3

1

58

Philip Turner

@philipturnerar

4 months

How to compile a logic rod in 50 lines of code. The data structure can be used to set up simulations on macOS or Windows.

2

57

Philip Turner

@philipturnerar

8 months

Not yet a reproduction of Eric’s knobs and 3D stacking, but close. Future projects will need exact reproductions of his logic and non-volatile RAM designs from 1992. [3/n]

2

0

54

Philip Turner

@philipturnerar

8 months

How to procedurally generate adamantane, a building block of many mechanosynthesis tooltips. #APM #MNT #nano

3

5

56

Philip Turner

@philipturnerar

8 months

Size is not the only benefit. The greatest one may be diversity. Macroscale glob, smash, stick manufacturing doesn't allow very intricate atomic scale patterns. With atomic precision, a lot of exotic hardware utilizing fragile quantum states could be viable. Besides QC. [8/n]

1

51

Philip Turner

@philipturnerar

7 months

The above was filmed on an Apple silicon laptop, at 120 FPS. [2/19]

1

0

51

Philip Turner

@philipturnerar

8 months

The simulated machinery was 16788 atoms (~6k for NOR gate, ~10k not strictly needed). For comparison, a TSMC N3 transistor can reach millions or billions of atoms, depending on whether you include 3D stacking inefficiency. [5/n]

1

49

Philip Turner

@philipturnerar

8 months

@LukeFarritor All of this is currently and will be open-source. Something of this caliber requires open information sharing from numerous people, and no delays from prioprietaryness. I want to make this a reality ASAP. It's been 35 years since Engines of Creation. Wanna wait 35 more?

1

50

Philip Turner

@philipturnerar

1 year

Reproducing Eric Drexler’s 1991 thesis, chapter 9: "Nanoscale Structural Components" #nanotechnology #diamondoid #APM

5

6

48

Philip Turner

@philipturnerar

1 year

@teortaxesTex 15% improvement is marginal. My GitHub release for Metal FlashAttention v1.0.0 measured the speedup in units of "order of magnitude".

3

1

48

Philip Turner

@philipturnerar

8 months

Real nano-technology blurs the boundary between software and hardware. Whether it "was built" in silico, or in experiment, in polymer chains compressed to self-generate AFM build sequences, etc. The era where matter becomes fully digital, hackable, accessible to the masses.

Ryan Orban

@ryanorban

8 months

Think of it: an ultra low-power, adaptive model trained on the World’s knowledge, uniquely tailored to each individual, weaving its lessons into a tapestry of engaging stories. Stepping into the pages of science fiction. What a time to be alive.

0

1

2

46

Philip Turner

@philipturnerar

8 months

Why would you get a 3090? Build a nanofactory and produce compute hardware for $1/kilogram. That would be 3200 GPUs for the same price as an M3 Max?

anton

@abacaj

8 months

If you’re doing inference locally, I would definitely just get a 3090. Why would you get a mac to do inference? The starting price of the m3 max is $3200 with 400gb/s bandwidth

61

26

664

2

47

Philip Turner

@philipturnerar

7 months

Second, shrink the design space. Only use 3 elements: C, Si, Ge. Avoid curved and rotary structures that require a GUI to place atoms, or exotic surface dopants (N, O, S) that don’t fundamentally alter functionality. This dodges a combinatorial explosion of geometries. [5/19]

1

0

45

Philip Turner

@philipturnerar

4 months

This is an energy minimization at the "rigid body mechanics" level of theory. [1/4]

7

2

45

Philip Turner

@philipturnerar

8 months

It may not sustain as high of clock speeds (500 MHz vs 5 GHz), or power efficiency, but the sheer compactness required makes it the best choice. However, factories that can build atomically precise machines can also build smaller electronics than TSMC. [7/n]

3

2

43

Philip Turner

@philipturnerar

1 year

Just casually outperforming Apple Accelerate and Metal Performance Shaders. cc @liuliu

5

42

Philip Turner

@philipturnerar

9 months

Compiler auto-vectorization can only do so much. The way I use Swift now, is very similar to how much of the explicitly vectorized Mojo code looks. I stick to 32-bit lanes with 8-wide vectors, regardless of CPU architecture. In Swift you can't set vector width to a variable.

Modular

@Modular

9 months

We're not biased at all 😏 but we believe Mojo 🔥 is what AGI gets built with 🚀😱💯

10

11

192

1

4

44

Philip Turner

@philipturnerar

1 year

I think I can make Stable Diffusion run 3x faster than Auto1111 WebUI, while consuming the same amount of power. Simulation of SD v1 UNet, 30 steps: - Metal Performance Shaders: 15.3 seconds - Metal FlashAttention: 5.0 seconds - Theoretical Lower Bound: 3.0 seconds

4

2

42

Philip Turner

@philipturnerar

9 months

@TentacleyShoko You use stiff covalent bonds instead of hydrogen bonds to hold it together. It's not a fun mental exercise, it's a practical manufacturing process, just requires a really expensive AFM.

1

0

43

Philip Turner

@philipturnerar

8 months

Regarding yield rate, it's a lot higher than many pessimists would like you to believe. But it is finite, so one would employ redundancy and fault tolerance. Build larger stuff with something akin to multi-die chips: convergent assembly. [9/n]

1

0

40

Philip Turner

@philipturnerar

9 months

61 trillion floating point operations per second, thanks to @liuliu ! Now I can simulate productive nanosystems with hundreds of thousands of atoms. This should satisfy the exponential scaling of my CAD capabilities for some time. End goal: 1 million atoms. [1/n] #MNT #APM #nano

3

1

37

Philip Turner

@philipturnerar

3 months

The story continues: atom by atom, row by row, layer by layer...

Philip Turner

@philipturnerar

3 months

Partially-hydrogenated partially-chlorinated Si(111). There's a story to go along with this image. [1/5]

1

0

21

1

39

Philip Turner

@philipturnerar

7 months

@MikePFrank I built a computer in Minecraft. A 64-bit adder.

2

0

39

Philip Turner

@philipturnerar

7 months

Third, use a DSL and geometry compiler to automate atom placement. 24 unique crystolecules were designed in 3 days. Crystolecules are large 3D molecules whose shape is defined by crystal planes. They contain sp3-bonded networks, in contrast to sp2-bonded carbon nanotubes. [6/19]

2

0

39

Philip Turner

@philipturnerar

11 months

Just learned what e/acc truly is - a movement to counter the AI/nuclear/nano doomerism from the likes of LessWrong, to show we have an optimistic future. It is NOT effective altruism.

2

4

36

Philip Turner

@philipturnerar

9 months

In Nanosystems, Eric laid out a lot of compiler technology that needs to be created, to get molecular manufacturing. This is the part that needs extreme automation, beyond what a human can design in just a GUI-based CAD like SOLIDWORKS.

4

2

34

Philip Turner

@philipturnerar

10 months

Successful compilation of a short domain-specific language excerpt into atoms #apm #nano

1

3

36

Philip Turner

@philipturnerar

7 months

First, a new approach to ray tracing. Build the acceleration structure once, render many times with zero cost per atom. The previous approach rebuilt the structure every frame, so geometry could change dynamically during an animation. [4/19]

1

0

35

Philip Turner

@philipturnerar

2 months

Going from M1, M2, M3, M4, the transistor count has increased exactly by 4 billion per year. Next chip will have 32 billion, then 36, then 40. We're in the era where transistor count scales linearly, not exponentially. Just like clock speed when Dennard Scaling stopped.

1

3

34

Philip Turner

@philipturnerar

6 months

A real scientific discovery made on a supercomputer powered by AMD GPUs. LFG.

Future Timeline

@future_timeline

6 months

Neutron star cores revealed by supercomputer One of the world's most powerful #supercomputers has simulated the largest and densest neutron stars. This new model shows there is a 90% chance they contain quark-matter cores. ▶️ Read more: #neutronstars

1

3

19

0

2

34

Philip Turner

@philipturnerar

7 months

I spent the last 6 days straight working on this project. Started with the goal to make something w/ 20M atoms, to prove I had fixed some long-standing bottlenecks with designing large structures. The intent is to design a massive nanomechanical ALU, then simulate it.

James Heaney

@jvheaney

7 months

@philipturnerar This work is awesome man, I really appreciate you sharing it! Looking forward to seeing more.

0

1

2

1

32

Philip Turner

@philipturnerar

8 months

I've learned this lesson multiple times. The more simple and minimal your codebase is, the more chances you find for drastic optimizations.

the tiny corp

@__tinygrad__

8 months

tinygrad's dependencies (all of them) "numpy", "requests", "tqdm", "networkx", "pyopencl"

23

22

672

1

32

Philip Turner

@philipturnerar

9 months

What I've learned, from a year researching how to make the smallest computer possible. You can make a computer out of anything. Silicon, DNA, quantum dots, proteins, diamond. Out of all, diamond mechanical is the most compact. Why? De Broglie wavelength of nuclei < electrons.

Philip Turner

@philipturnerar

9 months

You can build entire mechanical CPU systems out of just sp3 carbon and hydrogen (diamond, lonsdaleite). This is one of the few MNT products that is this way. Others like molecular manufacturing systems would benefit from more elemental diversity.

0

8

1

31

Philip Turner

@philipturnerar

1 year

@realGeorgeHotz tokens/s on ChatGPT-4 to validate my hypothesis? Another interesting part was 16 inferences. If OpenAI has each model inference, then converge, then scatter the thoughts back for a second inference, that’s massive. A new way of designing LLMs, far from “running out [2/n]

6

1

32

Philip Turner

@philipturnerar

7 months

Thus a nanosystem was designed in 3 days, using good software engineering practices. Restrict the design space, choose geometry that can be automated, sacrifice detail to save time. Organize atoms into data structures with mutable value semantics. [18/19]

1

32

Philip Turner

@philipturnerar

7 months

Worth noting why large structures were broken into smaller blocks. It was not for feasibility of manufacturing (smaller parts = higher yield). It was for **feasibility of design**. If the blocks are identical, it decreases compilation time. Compile once, copy many times. [7/19]

1

0

31

Philip Turner

@philipturnerar

4 months

Time to simulate some nanomachines.

3

1

31

Philip Turner

@philipturnerar

7 months

Germanium-substituted adamantane thiol is a tripod structure that can attach many different reactants. It can be used to build silicon atom-by-atom today. Carbon requires 2 scanning probes operated simultaneously, depicted by 2 tripods per robot arm. [10/19]

1

0

30

Philip Turner

@philipturnerar

8 months

@MikePFrank @kanzure The process is called second-generation high throughput AFM. You only need thousands of rods to build a replicator. Then, after resolving enough technical challenges, you can build sextillions of these in precise arrangement.

3

0

29

Philip Turner

@philipturnerar

11 months

If anyone is interested in how I learned GPGPU, the hard way. I wanted CUDA but couldn't access an NVIDIA GPU. So I wrote GPU algorithms from scratch using custom software, as no pre-existing (PyTorch, AMBER, etc.) optimized for Apple. Hardware needs to be hackable. @Modular_AI

Philip Turner

@philipturnerar

11 months

@mechadense @pplank8 Generally I think about the two major OS's that average people use - Mac and Windows. Linux is globbed together with Windows because most industry or HPC software (CUDA) runs on Linux/Windows. None of it runs on Mac/iPhone. So I'm reversing the industry paradigm a little.

1

0

3

0

3

27

Philip Turner

@philipturnerar

7 months

Nanomechanics doesn’t require a specialized CAD UI or Mech. Eng. degree (although they would help). If you know comp sci and good SWE practices, you can make major contributions. The backbone of the compiler is Swift, a modern programming language. [8/19]

Swift.org

Swift is a general-purpose programming language built using a modern approach to safety, performance, and software design patterns.

www.swift.org

1

0

29

Philip Turner

@philipturnerar

7 months

1,026,436 atoms, 1920 x 1920 resolution, 120.00 FPS. Why do you think I always include this watermark in my renders?

Philip Turner

@philipturnerar

7 months

@StephaneRedon @mooreth42 Apple ProMotion displays with 120 Hz are the best.

0

2

1

27

Philip Turner

@philipturnerar

8 months

@LukeFarritor CPU, registers, cache, and two tiers of RAM with different serial access speeds, but the same underlying technology.

6

0

28

Philip Turner

@philipturnerar

1 year

I'm going to push performance up to the limit permitted by physical law.

1

0

27

Philip Turner

@philipturnerar

7 months

Let’s dive into the convergent assembly architecture—how one builds atom-by-atom with parallel processing. [9/19]

2

0

28

Philip Turner

@philipturnerar

8 months

No, we're not butting heads with the laws of physics. We're approximately 785x away from the Landauer limit. It's the opposite: **Moore's Law is too slow.** It needs to run much, much faster.

Artillain

@artillain

8 months

@philipturnerar The problem is not making the transistors smaller, it's heat density, gate leakage, and other signs that we're starting to butt heads with the laws of physics and/or the materials currently used.

0

2

26

Philip Turner

@philipturnerar

8 months

I wonder, how many atoms does it take to build an artificial self-replicator? Start with the brain, the microprocessor.

6

0

23

Philip Turner

@philipturnerar

10 months

Strength of vdW adhesion in a stack of rhombic dodecahedra. 34k atoms, 7 x 80 ps. RT if you get the Minecraft reference. #APM #nano

6

3

26

Philip Turner

@philipturnerar

7 months

To add some color, housing was built from germanium. Small building blocks have pairs of (110) planes that form strong vdW bonds. They connect with (110) planes on the adjacent block, starting a chain. At the highest level, 4 chains interlock to hold the entire factory. [17/19]

1

27

Philip Turner

@philipturnerar

1 year

Why is my code faster than Metal Performance Shaders? Apple uses the hidden SIMD futures instructions in a way that needlessly zero-pads data fetched from device -> threadgroup memory. The farther you are from the nearest block size, the more severe this gets. I fixed that.

1

25

Philip Turner

@philipturnerar

4 months

Phosphorus.

3

2

25

Philip Turner

@philipturnerar

7 months

H, C, Si, S, Ge are all you need to build this.

Philip Turner

@philipturnerar

8 months

Atomically precise NOR gate, welcome to the world of computing. Any combinational logic circuit can be built with you. [1/n] #MNT #APM #nano

53

238

2K

2

1

25

Philip Turner

@philipturnerar

7 months

The third level is a robot arm with many degrees of freedom. This is a standard design taken straight from macroscale robotics. Two interlocked gears (not shown) drive opposing grippers. Although under a million atoms, it dwarfs the extremely compact rod logic structure. [15/19]

1

0

25

Philip Turner

@philipturnerar

11 months

Graphs of position and velocity over time. The housing has windows for viewing what’s going on inside. That creates a local potential maximum; the windows have less vdW binding energy than solid walls. The rod often stops and turns back halfway, requiring careful tuning. [2/11]

3

0

25

Philip Turner

@philipturnerar

1 year

In optimization, people often think of compute-bound and memory-bound. There’s another important one: latency-bound. Even the smallest, trivial operations take a non-trivial amount of time to finish. That is one major reason for operator fusion.

3

2

22

Philip Turner

@philipturnerar

7 months

Rod logic provides a compact, low-power method to transmit signals at the nanoscale. Three robot arms in the assembly line are addressed by 6 different logic rods. The stack of green dodecagons depicts a rotary machine that converts rod motion into arm motion. [11/19]

2

1

25

Philip Turner

@philipturnerar

3 months

A typical example of cubic diamond's natural tendency to warp. Fixing this requires precise modifications to change the orientation of stresses at the crystal surface.

4

1

25

Philip Turner

@philipturnerar

8 months

To get to type II you need atomic precision. Too much planetary material to not build in the most efficient way possible.

Pablo Moncada-Larrotiz

@pablo_moncada_

8 months

Got steaks and diet cokes with @BasedBeffJezos . Absolute legend. His vision for AGI is bigger than I could have imagined. He has type II Kardashev scale level ambitions, and a plan to get there. Incredibly excited for what’s to come.

10

4

125

2

0

22

Philip Turner

@philipturnerar

11 months

I’d like to thank @mechadense for some feedback on earlier prototypes. The files are hosted at under CC BY-SA 4.0. Source code is at . [11/11]

GitHub - philipturner/molecular-renderer: CAD program for molecular nanotechnology

CAD program for molecular nanotechnology. Contribute to philipturner/molecular-renderer development by creating an account on GitHub.

github.com

0

1

24

Philip Turner

@philipturnerar

11 months

The idea originated from . Use vdW forces to suction a rod into a tube. The first design had too large of a rod. Surfaces were too close, in the repulsive regime of the vdW potential. The housing rejected the rod. [3/11]

Lukas Süss

@mechadense

1 year

"VdW force wiggler concept" for better understanding of the expectable behavior of diamondoid crystolecules sliding freely on each other. The sine motion shown here is actually incorrect as …

1

4

2

0

23

Philip Turner

@philipturnerar

1 month

I built something, although it isn't diamond (yet).

3

2

23

Philip Turner

@philipturnerar

10 months

Adding the final pieces to a 10-part spring system (100 m/s), then watching it survive high-speed rotation (13 GHz, 400 m/s). 25k atoms. More renders may come in the future regarding this nanosystem. #mnt #apm #nano

5

4

22

Philip Turner

@philipturnerar

3 months

Nanosystems Chapter 14.6: Matter Compilers. That's a thing now.

0

3

23

Philip Turner

@philipturnerar

7 months

The first-level products are selectively passivated. Some crystal faces have no hydrogen atoms, and are free radicals. They readily “bond on contact” with another surface covered in radicals. The second level of hierarchical assembly is seamless covalent welding. [14/19]

1

0

23

Philip Turner

@philipturnerar

8 months

How about just bypass the wait and make this civilization right now? "We need to wait until 2100" not if atomic precision exists.

Beff – e/acc

@BasedBeffJezos

8 months

e/acc is the cultural framework of a more advanced civilization assembling itself from the future.

14

22

217

6

1

23

Philip Turner

@philipturnerar

4 months

sliding rod based computations... (correctly functioning 4-bit RCA)

Alberto Hojel

@AlbyHojel

5 months

@chesterzelaya @philipturnerar i think you know a two or thing abt this

0

2

1

22

Philip Turner

@philipturnerar

3 months

Partially-hydrogenated partially-chlorinated Si(111). There's a story to go along with this image. [1/5]

1

0

21

Philip Turner

@philipturnerar

1 year

Both the GEMM and FlashAttention shader are currently ~1000 lines. Partially because of all the constant folding, and the documentation of each minute elision of computations.

0

22

Philip Turner

@philipturnerar

8 months

@s_r_constantin This is a simulation, with the end goal of discovering atomically precise geometries that are easier to manufacture. I do intend to build this IRL within my lifetime, no joking.

2

1

21

Philip Turner

@philipturnerar

7 months

A belt of interlocking crystolecules captures the outputs of each assembly line. [13/19]

1

0

21

Philip Turner

@philipturnerar

3 months

5-membered rings are extremely useful for modulation of lattice warping. @StephaneRedon up to the challenge of reproducing this 2-part system in SAMSON? Image 1: after surface energy minimization / compilation Image 2: after full energy minimization with procedure in Image 3

Stephane Redon

@StephaneRedon

3 months

@philipturnerar This script is probably sufficient for that design stage: we can already deduce such structures are warping. A few weirdly placed hydrogens (if any) on edges don't seem to matter, and there doesn't seem to be a proof for a need of 'alternating patterns of 5-membered rings'.

0

2

0

20

Philip Turner

@philipturnerar

1 year

The news about room temperature superconductors is very, very exciting. If it ends up legit, it might be worth revisiting the common assumption that molecular assemblers are impossible too.

4

1

20

Philip Turner

@philipturnerar

7 months

Tripod tips equipped with reactive feedstocks, energy-minimized using GFN-xTB. These should theoretically insert SiH2 into an H-Si bond in a single step.

1

3

19

Philip Turner

@philipturnerar

7 months

The arm’s backbone consists of interlocking hexagons. Connected by strong vdW forces and stacked in an alternating pattern, they form a dense 3D housing structure. A hexagon can be designed in minutes, compiled once, and instantiated hundreds of times. [16/19]

1

0

21

Philip Turner

@philipturnerar

11 months

This is how superlubricity works. Two incompatible surfaces - the different spatial frequencies prevent snapping together in the direction orthogonal to motion. Hypothesized by Drexler (1991, bearings in Nanosystems), proven experimentally in 2004. [8/11]

Superlubricity - Wikipedia

en.wikipedia.org

1

0

20

Philip Turner

@philipturnerar

24 days

Setting up some tooltips for diamondoid mechanosynthesis simulations. Made possible by automated caching of relaxed structures and molecular dynamics trajectories. #MNT #APM #nano

3

1

24