Author: Philip Turner
Date Started: 3/3/2025
Date Completed: 3/6/2025
Abstract: Estimating the cost to build a diamondoid replicator
Table of Contents:
- Background
- Overview
- Terminology
- Cost Metrics
- Parameters
- Usage
- Example Calculation
- Conclusion
- Next Steps
The Drake equation calculates the theoretical number of alien civilizations in the Milky Way. It takes several factors, which are either known or have a range of uncertainty. Then, it multiplies them together to get an estimate. This approach removes the subjectiveness of estimates, letting the numbers speak for themselves. The answer comes from first principles.
| Parameter | Original Estimate | Current Estimate |
|---|---|---|
R_* |
1 yr-1 | 1.5–3 yr-1 |
f_p |
0.2–0.5 | 1 |
n_e |
1–5 | 0.2 |
f_1 |
1 | 0.00005–0.13 |
f_i |
1 | 10-9–1 |
f_c |
0.1–0.2 | 0.2 |
L |
103–108 yr | 304–109 yr |
N = R_* · f_p · n_e · f_1 · f_i · f_c · L
N = 20 [original lower bound]
N = 5 * 10^7 [original upper bound]
N = 9.1 * 10^{-13} [current lower bound]
N = 1.6 * 10^7 [current upper bound]
This equation inspired the following investigation.
I will estimate the cost to bootstrap molecular nanotechnology, as described in Nanosystems (1992). While the book describes a pathway with multiple steps (incremental path), I jump directly to the end goal. Scanning probe technology is the only proven technique for building atoms. There is no certainty that protein structures can even be controlled with 10-pm granularity yet 100-nm range. Only lithium niobate (macroscale) and diamondoid machines (nanoscale) can build atoms, based on the known design space.
Objective: Use lithium niobate 3DOF actuators to build semiconductor crystals (diamond, SiC, silicon) that can themselves build the same material. Understanding the cost of such an endeavor will guide experimental efforts to actually do it. The cost metrics will be specific and quantitative. For example, the minimum atom count for a replicating machine must be known.
Approach: Develop formulas for the four metrics listed below. After defining the correct relationships between parameters, list known parameter values (cite the source for each). State the range of uncertainty in each parameter. Multiply the smallest or largest value of each parameter. The product is the lower or upper bound of the answer.
- Time delay to achieve a functioning replicator
- Maximum acceptable upper bound: 5 years
- Spatial volume required for the laboratory
- Maximum acceptable upper bound: one cubic kilometer
- Financial cost (upfront)
- Maximum acceptable upper bound: 100% of US GDP
- Financial cost (ongoing)
- Maximum acceptable upper bound: 30% of US GDP
- Maximum acceptable upper bound: 90% of US annual helium production
The above limits will be assumed without justification. They serve as guiderails to permit the next step of this project, and can be refined in the future. The actual cost will ideally be much smaller.
The scope of this analysis does not include reaction success rates. We will assume that errors can be corrected, minimized by switching to cryogenic temperatures, or imperfect products are tolerable. Too little is known to connect proven SPM failure rates (10-1) to theoretical estimates (10-15). Such numbers would be more well understood with experimental validation that diamond can be built. This document resolves a key problem standing in the way of setting up such an experiment. Namely, that we have no rigorous assessment of how long/how much upfront investment it will take to return a profit.
Hypothesis: Commercial SPMs, overpriced and bottlenecked by piezoelectric creep, make bootstrapping economically unfeasible. Custom hardware based on lithium niobate makes bootstrapping at least 103 times more likely. In addition (not part of this analysis), lithium niobate will improve the success rates, because it makes nanopositioning more accurate. Specifically, nanopositioning becomes linear time-invariant for the first time in history.
atom builder
| (≥100 nm)^3 build volume
| 14 bits of true accuracy
| 100 nm / 2^14 = 10 pm granularity
|
+ ---> scanning probe microscope (SPM)
| reliable, repeatable 3DOF nanopositioning
|
+ ---> diamondoid replicator
reliable ≥3DOF nanopositioner fitting inside (≤1 µm)^3
Assumption: Linear (analog) signals cannot be handled by nanoscale electrical interconnects. Only digital (1-bit) serial signals. Data must be transmitted from macroscale DSPs through digital electrical signals. At the replicator's I/O interface, data converts from electrical to mechanical. It is decoded into 14-bit parallel words with compact rod logic.
A replicator must satisfy the following constraints:
- Form closure
- Encapsulation of three 14-bit resolution linear actuators
- At least 3 degrees of freedom
- Vacuum seal (if operating under UHV conditions, 10-9 torr)
- Must be atomically precise, to have a defined, tightly sealed valve for transferring feedstocks in and out
- Ability to detect errors (if 10-5 error rate)
- Requires rod logic computing
- Requires a transimpedance amplifier or alternative to the qPlus sensor, with orders of magnitude smaller dimensions
It does not need to satisfy:
- Encapsulation of digital signal processing (DSP) hardware
- Vacuum seal (if operating under XHV conditions, 10-13 torr and no leakage when transferring between vacuum chambers)
- Ability to detect errors (if 10-15 error rate)
List of replicators and ≥3DOF actuators:
| Atom Count | Description | Source |
|---|---|---|
| 23 | Conventional mode mechanosynthesis tip, only has form closure | 1 |
| 21,318 | 3-DOF flexure, but lacks the 14-bit stepper motors | 2 |
| 4,000,000 | 6-DOF robot arm with 86 nm range, unspecified resolution | Nanosystems, Ch. 13 |
| 203,908,873 | 12-DOF pair of Stewart platforms, full vacuum enclosure | KSRM |
A replicator's operation frequency is expected to be in the MHz regime. It is an open question whether a nanoscale atom builder needs to utilize convergent assembly to bypass throughput bottlenecks. If so, it wouldn't be a monolithic replicator, but more like a factory. At the macroscale, convergent assembly is definitely needed. Operation frequencies are kHz or less.
List of scanning probe microscopes:
| Operation Speed | Description | Source |
|---|---|---|
| 0.016 Hz | Piezo creep settling time (60 seconds) | 1 |
| 1 Hz | qPlus AFM frequency shift, hyper-resolution images of organic molecules, 4 K | 2 |
| 98 Hz | qPlus AFM frequency shift, fuzzy images, NiO(001), 300 K | 2 |
| 190 Hz | qPlus AFM frequency shift, very fuzzy images, KBr, 300 K | 2 |
| 1635 Hz | Lowest eigenmode of UHV-STM | 3 |
| 5400 Hz | Lowest eigenmode of Air-STM/EC-STM | 4 |
The greatest unknowns reside in component 2:
- Not accurately estimating the number of design iterations needed. If one can only make one shot at a replicator, and it takes multiple decades to build, one will surely fail. Nothing works the first time. The reason it previously seemed impossible: we previously could not fathom having enough breathing room to attempt multiple times.
- Not accurately estimating the parallelism required. For example, bulk chemistry (DNA origami) has been proposed because it can make 1012 atom builders in parallel. However, it is questionable whether DNA origami can even achieve linear time invariance with 14 bits of resolution. A similar argument holds for MEMS scanning probes.
- Regarding parallelism, each probe must be independent. If one probe undergoes a failed reaction, all the other probes cannot be stopped simultaneously. Each probe must be able to adapt to its failed reaction, independently of the others. This is an issue with ideas for MEMS scanning probes with SIMD architectures (64 control signals controlling 1024 probes on a chip).
- Another issue with fully integrated MEMS SPMs, would be the flexibility of swapping samples and tips. Furthermore, it is questionable whether the coarse approach mechanism could be entirely integrated into MEMS. If not, there is no benefit to using MEMS. The dominant contributor to resonance frequency comes from the compliance of the coarse approach mechanism. MEMS is supposed to be an optimization that improves the true resonance frequency.
Non-multiplicative relationships between variables:
- We don't know whether parts from failed design iterations can be recycled in subsequent ones. If so, the relationship between
atoms/replicatorand(design iterations)/(successful replicator)is non-multiplicative. To stay conservative, we will not elaborate on this possibility. - If the number of scanning probes is too large, we may suffer from Amdahl's law. A replicating system consists of multiple parts, which will be mated through nonbonded (e.g. vdW, magnetic) interactions. If a single SPM cannot produce a single part in the required amount of time, there is a serial bottleneck. This is mostly a concern if piezoelectric creep cannot be solved.
component 1: work to build one replicator, in (scanning probe)-hours
feedback loop latency * operations/reaction * reactions/atom * atoms/replicator
component 2: time to build a functioning replicator
component 1 * (design iterations)/(successful replicator)
* 1 / (#scanning probes working in parallel)
This metric serves to plan how much 2D area to allocate, before building a semiconductor fab. Modern semiconductor fabs have multiple machines working in parallel. They facilitate different steps of the wafer production process, such as cleaning and depositing interconnects. It is an optimized, pipelined workflow, just like a modern out-of-order CPU core. Regarding the extreme ultraviolet lithography (EUV) step, a brief internet search suggests 10 EUV machines per fab.
For nanotechnology, there are multiple steps:
- Sample preparation
- Potentially molecular beam epitaxy
- Potentially electrochemical preparation of silver surfaces
- Potentially XPS analysis to verify correct preparation
- Tip preparation
- Using an e-beam or argon ion beam to remove the oxide layer from a silicon probe
- Using the coarse Z approach actuator to approach the probe to the surface with tripods
- Inverted mode mechanosynthesis
- Using the coarse XY actuator to switch to a different tripod located on the surface (they must be very sparse, separated by a distance much larger than the silicon probe's radius)
- Using STM feedback (large tunneling current from the silver surface with alcohol linkers, hopefully) to align with the Si lattice unit cell on the silicon probe
- Identify the pre-planned location for atom deposition, with unspecified, efficient computer vision techniques
- Using a pair of orthogonal, 1D scanlines to quickly verify reaction success. The computer vision algorithm does not need 2D images like in typical scanning probe microscopy. It is necessary to reduce the cost of imaging from
$n^2$ to$2n$ , where$n$ is the feature width divided by the pixel width.
The hardware for each of the three steps may require separate vacuum chambers. In addition, there may be "highways" to transport manufactured parts between chambers. The highways would begin at a vacuum chamber with a scanning probe, and end at a central chamber for manipulation with a coarse scanning electron microscope (SEM). The central chamber would assemble the incoming line of parts into a functioning system. This is convergent assembly, except done in a macroscale factory instead of a nanoscale factory.
Non-multiplicative relationships between variables:
- The highways connecting vacuum chambers to each other. We will omit them from the calculation, for brevity. Interconnects in high-performance computing are often a performance bottleneck. In my experience with rod logic, gates had to be laid out in a spatial arrangement that favors data transmission. This constraint may have exploded the volume by 3x.
- Model the XPS and MBE systems (if used at all) as existing in discrete vacuum chambers. The equation does not explicitly model the optimization that fuses them both into the same chamber.
- Space occupied by gas liquefaction systems (if used at all) is considered negligible.
component 1: 2D footprint of a vacuum chamber
(chamber dimension + spacing between chambers)^2
component 2: number of SPM chambers
(#scanning probes working in parallel) / (scanning probes/SPM chamber)
component 3: number of overall vacuum chambers
number of SPM chambers *
(1 + XPS systems/SPM chamber + MBE systems/SPM chamber)
component 4: amount of 2D area to allocate
component 1 * component 3 / (number of building storeys)
Vacuum chambers operating at 4 K cannot have windows/viewports. Unsure whether vacuum chambers operating at 77 K can. To be conservative, assume viewports are only allowed when the temperature is 300 K. However, to reduce the number of non-multiplicative relationships, include the cost of viewports in all estimates. Even if cryogenic temperatures are required.
The cost of occupying land (upfront and ongoing) is not included in estimates, to simplify them.
component 1: cost per vacuum chamber
cost of roughing pump +
cost of turbo pump +
cost of ion pump +
cost of vacuum chamber walls +
(cost of CF flange) * (flange count) +
(cost of viewport) * (viewpport count)
component 2: cost per SPM chamber
component 1 +
component 1 * (XPS systems/SPM chamber) +
component 1 * (MBE systems/SPM chamber) +
(cost of piezo driver) * (scanning probes/SPM chamber) +
(cost of piezos) * (scanning probes/SPM chamber) +
(cost of gas liquefaction system) * (liquefaction systems/SPM chamber)
component 3: total cost
cost per SPM chamber * number of SPM chambers
component 4: barrier to entry, for a single mechanosynthesis experiment at 300 K
compoment 1 +
component 1 * (using XPS ? 1 : 0) +
component 1 * (using MBE ? 1 : 0) +
cost of piezo driver +
cost of piezos +
cost to synthesize tripods
Assumption: Machines operate at 100% duty cycle, with minimal human intervention. The chemistry must support automation of lattice construction. For example, either silicon carbide or carbon nanotubes.
Liquid neon is more expensive than liquid helium, per liter. LNe would only be used if we're already employing an on-site gas liquefaction system. Neon might reduce the power consumption or leak rate, thus reducing the cost of the liquefaction system.
If LHe is not recycled, implicitly assume that each SPM chamber has a dedicated cryostat. There is no parameter for a ratio of continuous flow cryostats or bath cryostats to SPM chambers. Note that this situation differs from the case for liquefaction systems, where the ratio could plausibly be reduced.
Amount of helium produced in the US:
- 79 million cubic meters sold/year (1)
- 1.86 billion cubic meters left in stockpile (1)
- 8.49 billion cubic meters in known geologic natural gas reservoirs (1)
component 1: cost of cryocoolant, per (SPM chamber)-hour
ambient | 300 K | 0
LN2 | 80 K | (cryostat liters/hour) * (cost of LN2/liter)
LNe, recycled | 30 K | 0
LH2 | 25 K | (cryostat liters/hour) * (cost to purify LH2/liter)
LHe | 10 K | (cryostat liters/hour) * (cost of LHe/liter)
LHe, recycled | 5 K | 0
component 2: power consumption, per SPM chamber
power of roughing pump +
power of turbo pump +
power of ion pump +
(power of piezo driver) * (scanning probes/SPM chamber) +
(power of gas liquefaction system) * (liquefaction systems/SPM chamber)
component 3: cost of power, per (SPM chamber)-hour
(power per SPM chamber) * (1 kW / 1000 W) *
(cost of power in Virginia, per kWh)
component 4: total hardware cost, per year
[cost of cryocoolant + cost of power], per (SPM chamber)-hour
* (number of SPM chambers) * (hours/year)
component 5: total person hours cost, per year
(number of employees) * (wage in dollars/hour) * (length of shift / 24 hours)
* (workdays/week) * (hours/year)
component 6: helium consumed, per year
(cryostat liters/hour) * (hours/year) * (number of SPM chambers)
* (liquid helium kg/liter) * (gaseous helium m^3/kg)
Atoms per replicator:
- [Lower bound] 4 million, 6-DOF actuator with no vacuum enclosure
- [Upper bound] 204 million, 12-DOF actuator with vacuum enclosure
Feedback loop latency:
- [Lower bound] 0.6 milliseconds, STM on conducting surfaces
- Images get blurry when scanline rate exceeds 1–2 kHz, even though piezo XY motion shouldn't couple to coarse approach mechanism's Z eigenmode)
- [Upper bound] 10 milliseconds, qPlus AFM with "good enough" resolution on insulating surfaces (98 Hz)
- qPlus frequency shift is not equal to sensor bandwidth, according to theory
- Actual imaging times (1 hour for hyper-resolution, 1 Hz/pixel) suggest it's at least proportional
- Not relying on SPM designs that necessitate 4 K (e.g. hyper-resolution qPlus AFM). Note that one objective of this document, is to understand what would happen if piezo creep can't be solved. The final estimates should illustrate what would happen, if latency is 1 second or 1 minute. Consequently, LHe is required, increasing the lower bound to the cost estimate.
Operations per reaction:
- [Lower bound] CBN reaction time / piezo creep settling time
- Reaction time = 8 hour workday / 6 atoms/day
- Settling time = 60 seconds (US12174218B2)
- Formula evaluates to 80 operations
- [Upper bound] Probe diameter / pixel size * 2 orthogonal directions (X, Y) * log2(probe diameter / pixel size)
- Probe diameter = 10 nm
- Pixel size = 1/4 of the Si(100) surface unit cell
- Formula evaluates to 600 operations
- Final term of upper bound constitutes a binary search algorithm
Probes per SPM chamber:
- [Lower bound] 4 has been achieved while reaching both UHV and 4.5 K (1)
- [Upper bound] 6, to be conservative
- This is a good place for future investigation, most likely quantification of the limiters regarding PCB interconnection to outside of the vacuum chamber. In addition, whether mechanical coupling between nearby SPMs harms vibration isolation.
- The physics/geometry of a cryostat that doesn't recycle the liquid helium, may necessitate only 1 scanning probe per chamber. We won't include this pessimistic outcome in the estimates.
Number of building storeys:
- [Lower bound] 1
- [Upper bound] 2, to be conservative
- Semiconductor fabs have different levels allocated to different functions. One might be a ventilation system for a clean room above. One larger clean room may be more economical than two smaller clean rooms.
Liquefaction systems per SPM chamber
- [Lower bound] 0.5, to be conservative
- [Upper bound] 1
- This represents an optimization opportunity. Since the upfront cost of a liquefaction system is so great, perhaps use one liquefaction system to cover multiple machines.
Design iterations to build a successful replicator
- [Lower bound] 11, number of Apollo missions to land on the Moon
- [Upper bound] 200, because 1000 seems too pessimistic
- Edison actually tried more than 1000 times to create the lightbulb: (1)
- There is significant room for improvement on this estimate. Particularly, on shrinking the upper bound.
| Build Sequence | Total Reactions | Total Group IV Atoms | Ratio |
|---|---|---|---|
| Diamond | 62 | 40 | 1.55 |
| Silicon | 48 | 13 | 3.69 |
| SiC, Inverted Mode | 90 | 23 | 3.91 |
Source 1: GitHub
Source 2: GitHub
Every time the vacuum chamber is opened, copper CF flanges may need to be replaced. In the worst case, the chamber must be baked out every 2 weeks. For simplicity, the time delay to a replicator is 2 years, even though the following number may be used for a scenario with 5 years. (2 years / 2 weeks) * (2 flanges/baking out) = 104 flanges used. Add this number to the lower bound for "flange count".
Costs of components: CF flanges (1), which appear UHV-grade. The only source for viewports, after a quick search, is ThorLabs. These appear HV-grade, not UHV-grade (2). From these sources, I can make crude estimates for some parts I'm not yet familiar with.
| Part | Lower Bound | Upper Bound |
|---|---|---|
| KF flange | negligible, not UHV quality | negligible, not UHV quality |
| CF flange | $50 | $100 |
| viewport | $300 | $300 |
For helium, use the most efficient cryostats possible, the bath cryostat. It is very difficult to use, but we have no choice. It is important to minimize ongoing costs. The only alternative is an on-site gas liquefaction system, which costs even more. Except... I cannot get a good number from a quick internet search. I will use 0.45 L/h, which likely corresponds to a continuous flow cryostat. These machines are easier to use than bath cryostats. In addition, bath cryostats have a large overhead to fill with the gas before taking an SPM image. This overhead could counteract the lower per-hour cost.
Here is the final decision:
| Coolant | Optimistic | Pessimistic |
|---|---|---|
| LHe | 0.164 L/h (1) | 0.5 L/h (2) |
| LN2 | 0.1 L/h (2) | 2 L/h |
The following steps must be performed, to acquire cost estimates.
Round 1:
- Fill in the parameters that contribute to "Time Delay".
- Budget either 2 or 5 years to build a functioning replicator.
- Figure out the number of SPMs required. There should be four numbers, lower/upper for the 2-year scenario and lower/upper for the 5-year scenario.
- Apply Amdahl's law.
- If SPMs are producing parts with under 10,000 atoms, extend the amount of time required.
- 204 million / 10k = 20400 SPMs maximum
Round 2:
- Fill in the parameters that contribute to "Space".
- Calculate the space and the number of vacuum chambers.
Round 3:
- Fill in all of the remaining parameters in the equations above. Some, such as those for components of a vacuum chamber, might be appropriate as tables.
- Calculate the financial cost of the entire effort, over all 2 or 5 years. Assume liquid helium is not needed.
Round 4:
- Repeat round 3 for scenarios that need cryogenic temperatures:
- [Scenario] Liquid nitrogen needed (consumed)
- [Scenario] Liquid helium needed (consumed)
- [Scenario] Liquid helium needed (recycled)
- Repeat rounds 1, 2, 3 for scenarios where atom placement is horribly slow
- Require liquid helium. Based on the previous analysis, go with the cheapest of "consumed" vs. "recycled".
- [Scenario] Hyper-resolution qPlus AFM needed
- [Scenario] Piezo creep cannot be solved
In Nanosystems, it is quoted that 90% of the atoms in a mechanical device are housing. To reduce the amount of uncertainty, we will increase the lower bound for replicator atom count. Assuming the 6DOF manipulator constitutes the 10% of atoms contributing to moving parts.
The instructions say to consider both lower/upper bound, for both 2 and 5 year scenarios. Instead, set the 2 year scenario as the optimistic bound, with 11 design iterations. The 5 year scenario is the pessimistic bound, with 200 design iterations. This choice reduces the range of uncertainty.
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| feedback loop latency | 0.6 ms | 10 ms |
| operations/reaction | 80 | 600 |
| reactions/atom | 1.55 | 3.91 |
| atoms/replicator | 40 million | 204 million |
| design iterations/successful replicator | 11 | 200 |
| time to build a functioning replicator | 2 years | 5 years |
component 2 = component 1 * (design iterations)/(successful replicator)
* 1 / (#scanning probes working in parallel)
#scanning probes working in parallel = component 1
* (design iterations)/(successful replicator) * (1 / component 2)
| Component | Optimistic | Pessimistic |
|---|---|---|
| component 1 | 2.98e6 SPM-seconds | 4.82e9 SPM-seconds |
| component 2 | 6.31e7 seconds | 1.58e8 seconds |
| #scanning probes working in parallel | 0.52 | 6,100 |
Repeating this analysis for the scenarios from "Round 4". First, hyper-resolution qPlus AFM. Although the images from the Geissbl paper had a frequency shift at 1 Hz, a plausible sensor bandwidth is 5 Hz. The worst-case scenario violated Amdahl's law, so I ran the calculations a second time with 30 years.
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| feedback loop latency | 0.2 s | 0.2 s |
| time to build a functioning replicator | 2 years | 30 years |
| Component | Optimistic | Pessimistic |
|---|---|---|
| component 1 | 9.92e8 SPM-seconds | 9.57e10 SPM-seconds |
| component 2 | 6.31e7 seconds | 9.46e8 seconds |
| #scanning probes working in parallel | 170 | 20,200 |
Finally, for the scenario bottlenecked by piezoelectric creep (CBN's current situation). Both of the bounds violated Amdahl's law. For the optimistic scenario, the limit is 40 million atoms / 10,000 atoms/part = 4000 parts. I ran the optimistic bound again, at 30 years. The pessimistic scenario must be 10,000 years.
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| feedback loop latency | 60 s | 60 s |
| time to build a functioning replicator | 30 years | 10,000 years |
| Component | Optimistic | Pessimistic |
|---|---|---|
| component 1 | 2.98e11 SPM-seconds | 3.38e13 SPM-seconds |
| component 2 | 9.46e8 seconds | 3.15e11 seconds |
| #scanning probes working in parallel | 3,470 | 21,500 |
The results of this round are summarized below. To simplify calculations in subsequent rounds, some significant figures were removed from estimates with very large absolute values.
| Time Delay | Optimistic | Pessimistic |
|---|---|---|
| Fast SPM Sensors | 2 years | 5 years |
| Hyper-Resolution qPlus AFM | 2 years | 30 years |
| Piezo Creep Bottleneck | 30 years | 10,000 years |
| SPMs in Parallel | Optimistic | Pessimistic |
|---|---|---|
| Fast SPM Sensors | 1 | 6,000 |
| Hyper-Resolution qPlus AFM | 170 | 20,000 |
| Piezo Creep Bottleneck | 3,500 | 20,000 |
The time delay constraint, is that the pessimistic estimate cannot exceed 5 years.
| Meets Time Delay Constraint | Optimistic | Pessimistic | Overall |
|---|---|---|---|
| Fast SPM Sensors | ✅ | ✅ | ✅ |
| Hyper-Resolution qPlus AFM | ✅ | ❌ | ❌ |
| Piezo Creep Bottleneck | ❌ | ❌ | ❌ |
To avoid errors, apply both the lower and upper bounds for the following parameters, to both the "Optimistic" and "Pessimistic" estimates from Round 1. The result should be four different numbers. At the end of the round, judge which two numbers to pick out of the four. The choice should maximize the gap between optimistic and pessimistic estimates.
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| chamber dimension | 0.5 m | 0.5 m |
| spacing between chambers | 2.0 m | 2.0 m |
| scanning probes/SPM chamber | 4 | 6 |
| XPS systems/SPM chamber | 0.25 | 0.5 |
| MBE systems/SPM chamber | 0.25 | 0.5 |
| number of building storeys | 1 | 2 |
For parameters that multiply in the formula, a pessimistic case has a larger value. For parameters that divide in the formula, a pessimistic case has a smaller value. This realization might resolve the ambiguity between lower/upper bounds and optimistic/pessimistic estimates.
| Parameter | Effect | Optimistic Case | Pessimistic Case |
|---|---|---|---|
| chamber dimension | multiply | 0.5 m | 0.5 m |
| spacing between chambers | multiply | 2.0 m | 2.0 m |
| scanning probes/SPM chamber | divide | 6 | 4 |
| XPS systems/SPM chamber | multiply | 0.25 | 0.5 |
| MBE systems/SPM chamber | multiply | 0.25 | 0.5 |
| number of building storeys | divide | 2 | 1 |
| Component | Optimistic | Pessimistic |
|---|---|---|
| component 1 | 6.25 m2 | 6.25 m2 |
| component 2 | #SPMs / 6 | #SPMs / 4 |
| component 3 | #SPMs * 0.25 | #SPMs * 0.5 |
| component 4 | #SPMs * 0.781 m2 | #SPMs * 3.13 m2 |
For the results of Round 1, we will calculate the following. #SPMs * 0.781 m2 for the optimistic case. #SPMs * 3.13 m2 for the pessimistic case. This choice appears to be the correct one. In addition, calculate the square side length of the facility. It is the square root of the area.
| 2D Area Allocated | Optimistic | Pessimistic |
|---|---|---|
| Fast SPM Sensors | 0.781 m2 | 18,800 m2 |
| Hyper-Resolution qPlus AFM | 133 m2 | 62,600 m2 |
| Piezo Creep Bottleneck | 2,730 m2 | 62,600 m2 |
| 1D Facility Dimension | Optimistic | Pessimistic |
|---|---|---|
| Fast SPM Sensors | 0.884 m | 137 m |
| Hyper-Resolution qPlus AFM | 11.5 m | 250 m |
| Piezo Creep Bottleneck | 52.2 m | 250 m |
The space constraint, is that the pessimistic estimate cannot exceed 1 kilometer in any dimension.
| Meets Space Constraint | Optimistic | Pessimistic | Overall |
|---|---|---|---|
| Fast SPM Sensors | ✅ | ✅ | ✅ |
| Hyper-Resolution qPlus AFM | ✅ | ✅ | ✅ |
| Piezo Creep Bottleneck | ✅ | ✅ | ✅ |
Costs for the pumps are recited from memory, without citing the sources. Similarly, the cost of vacuum chamber walls is set arbitrarily. For formulas with such a large number of parameters, we must cut corners to save time investigating. The effect on the final results is less severe; these numbers add, instead of multiplying together. More rigorous theoretical investigation could give very high-quality estimates for these costs.
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| cost of roughing pump | $2,000 | $2,000 |
| cost of turbo pump | $5,000 | $12,000 |
| cost of ion pump | $1,000 | $2,000 |
| cost of vacuum chamber walls | $8,000 | $8,000 |
| cost of CF flange | $50 | $100 |
| flange count | 10 | 114 |
| cost of viewport | $300 | $300 |
| viewport count | 3 | 3 |
Cost per vacuum chamber: $17,400 (optimistic), $36,300 (pessimistic). This is orders of magnitude lower than commercial systems, priced at over $1,000,000 (quoting Scienta Omicron's CEO) even for systems that might not have LHe. However, these costs don't include many auxiliary costs, such as vibration isolation and highways between chambers. They also don't include the person-hours cost to set up and maintain vacuum chambers.
We must be very efficient, about balancing the tradeoff of making chambers easy to use. For example, maglev turbopumps ($12,000) have more upfront cost, but less person-hours spent periodically changing the oil. In addition, less time spent re-outgassing the chamber after servicing the pump. Oil-lubricated pumps might be changed out, on the order of 5 times per year. Probably more than most commercial systems, as our facility will run machines at 100% duty cycle (24/7).
Another series of parameters, pulled with little justification and no sources cited. This is to save time getting an estimate out; further work can refine it. First, the case that requires XPS and MBE, but not cryogens. Alternatively, cryogens, but not recycled. Assume the cost of such a cryostat (e.g. continuous flow LHe cryostat) to be negligible.
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| XPS systems/SPM chamber | same as Round 2 | same as Round 2 |
| MBE systems/SPM chamber | same as Round 2 | same as Round 2 |
| cost of piezo driver | $300 | $300 |
| cost of piezos | $300 | $1500 |
| scanning probes/SPM chamber | same as Round 2 | same as Round 2 |
For the optimistic case, we actually increase the cost of the SPMs in each chamber.
| Parameter | Optimistic | Pessimistic |
|---|---|---|
| scanning probes/SPM chamber | 6 | 4 |
| Component | Optimistic | Pessimistic |
|---|---|---|
| component 1 | $17,400 | $36,300 |
| component 2 | $29,700 | $79,800 |
| component 3 | #SPMs * $4,950 | #SPMs * $20,000 |
| component 4 | $52,800 | $110,700 |
Second, the case that requires both cryogens and cryogen recycling. We are using numbers recited from commercial sources. I recall two distinct sources, one saying $60,000 and the other saying $70,000. There might be a high upfront cost to develop a custom design with lower cost. It would only be beneficial late-stage, to optimize the cost of the semiconductor fab. One optimization may come from switching to liquid neon (estimate: 2x cost reduction to fabricate a liquefaction system).
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| cost of gas liquefaction system | $60,000 | $70,000 |
| liquefaction systems/SPM chamber | 0.5 | 1 |
| Component | Optimistic | Pessimistic |
|---|---|---|
| component 2 | $59,700 | $150,000 |
| component 3 | #SPMs * $9,950 | #SPMs * $37,500 |
The pessimistic cost aligns with something I heard in private conversation, that an SPM with liquid helium recycling could be plausible for as low as $250,000. We don't have many numbers, from any sources, published or not.
Finally, combine the two cases (LHe recycling or not) with the SPM count. Ultra-slow imaging techniques require 4 K for stability. For the most optimistic case with fast SPM sensors, it is unrealistic to multiply cost/SPM by one. There is a non-multiplicative relationship, as the minimum number of SPMs/chamber is 6. However, there is yet another non-multiplicative relationship. The minimum number of XPS/MBE systems is 1. Therefore, we just state the lowest estimate for upfront cost to run a single mechanosynthesis experiment.
| Upfront Cost | Conditions | Optimistic | Pessimistic |
|---|---|---|---|
| Fast SPM Sensors | room temperature | $53,000 | $120 million |
| Fast SPM Sensors | LHe, recycled | $113,000 | $225 million |
| Hyper-Resolution qPlus AFM | LHe, not recycled | $842,000 | $400 million |
| Hyper-Resolution qPlus AFM | LHe, recycled | $1.7 million | $750 million |
| Piezo Creep Bottleneck | LHe, not recycled | $17 million | $400 million |
| Piezo Creep Bottleneck | LHe, recycled | $34 million | $750 million |
The upfront cost constraint, is that the pessimistic estimate cannot exceed 1 year of US trade volume. As of 2023, the US GDP was $27.7 trillion.
| Meets Upfront Cost Constraint | Conditions | Optimistic | Pessimistic | Overall |
|---|---|---|---|---|
| Fast SPM Sensors | room temperature | ✅ | ✅ | ✅ |
| Fast SPM Sensors | LHe, recycled | ✅ | ✅ | ✅ |
| Hyper-Resolution qPlus AFM | LHe, not recycled | ✅ | ✅ | ✅ |
| Hyper-Resolution qPlus AFM | LHe, recycled | ✅ | ✅ | ✅ |
| Piezo Creep Bottleneck | LHe, not recycled | ✅ | ✅ | ✅ |
| Piezo Creep Bottleneck | LHe, recycled | ✅ | ✅ | ✅ |
I am deviating from the instructions. This is the round where ongoing costs are determined.
$14.25 per kilogram of hydrogen at the pump (1)
| Gas Property | Value |
|---|---|
| LN2 density (kg/L) | 0.807 |
| LH2 density (kg/L) | 0.0709 |
| LHe density (kg/L) | 0.125 |
| gaseous N2 density (m3/kg) | 0.800 |
| gaseous H2 density (m3/kg) | 11.1 |
| gaseous He density (m3/kg) | 5.62 |
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| LN2 liters/hour | 0.1 | 2.0 |
| LH2 liters/hour | 2.0 | 2.0 |
| LHe liters/hour | 0.164 | 0.5 |
| LN2 cost/liter | $1.78 | $2.50 |
| LH2 cost/liter | $1.01 | $1.01 |
| LHe cost/liter | $20 | $50 |
I am using the following rule for person hours cost. Best case, we have 2 employees monitoring the factory at any moment. There are 8-hour shifts, and people monitor 24/7. We are paying someone to be there at every hour. Worst case, we have one employee per 10 SPM chambers. I don't have any reasonable estimate for hourly wage, so I'm throwing in $30–$50.
| Parameter | Lower Bound | Upper Bound |
|---|---|---|
| power of roughing pump | 300 W | 550 W |
| power of turbo pump | 100 W | 300 W |
| power of ion pump | 40 W | 200 W |
| power of piezo driver | 10 W | 50 W |
| power of gas liquefaction system | 500 W | 1000 W |
| energy cost per kWh | $0.14 | $0.14 |
| employees on guard, 24/7 | 2 | #SPM chambers / 10 |
| hourly wage | $30 | $50 |
| component 1 | Temperature | Optimistic | Pessimistic |
|---|---|---|---|
| ambient | 300 K | $0.00 | $0.00 |
| LN2 | 80 K | $0.18 | $5.00 |
| LNe, recycled | 30 K | $0.00 | $0.00 |
| LH2 | 25 K | $2.02 | $2.02 |
| LHe | 10 K | $3.28 | $25.00 |
| LHe, recycled | 5 K | $0.00 | $0.00 |
| component 2 | Optimistic | Pessimistic |
|---|---|---|
| not recycling a cryogen | 500 W | 1250 W |
| recycling a cryogen | 750 W | 2250 W |
| component 3 | Optimistic | Pessimistic |
|---|---|---|
| not recycling a cryogen | $0.070 | $0.175 |
| recycling a cryogen | $0.105 | $0.315 |
Remember that the number of SPM chambers is the number of SPMs, divided by 6 or 4. This division is noted in the table below, but not explicitly evaluated. This is to reduce the chance for errors, and make the calculations more traceable. Since this specific calculation is quite lengthy and error-prone.
| component 4 | Temperature | Optimistic | Pessimistic |
|---|---|---|---|
| ambient | 300 K | #SPMs * $613 / 6 | #SPMs * $1533 / 4 |
| LN2 | 80 K | #SPMs * $2190 / 6 | #SPMs * $45000 / 4 |
| LNe, recycled | 30 K | #SPMs * $920 / 6 | #SPMs * $2760 / 4 |
| LH2 | 25 K | #SPMs * $18300 / 6 | #SPMs * $19200 / 4 |
| LHe | 10 K | #SPMs * $29300 / 6 | #SPMs * $221000 / 4 |
| LHe, recycled | 5 K | #SPMs * $920 / 6 | #SPMs * $2760 / 4 |
component 5: revised formula for person hours cost, per year
(employees on guard, 24/7) * (wage in dollars/hour) * (hours/year)
| Component | Optimistic | Pessimistic |
|---|---|---|
| component 5 | $526000 | ((#SPMs / 4) / 10) * $438000 |
| component 6 | #SPMs * 168 m3 | #SPMs * 769 m3 |
Finally, we will evaluate all of the ongoing costs. First, cost of cooling solutions for the case of "Fast SPM Sensors".
| Ongoing Cost | Temperature | Optimistic | Pessimistic |
|---|---|---|---|
| ambient | 300 K | $613/yr | $2.3 million/yr |
| LN2 | 80 K | $2,190/yr | $68 million/yr |
| LNe, recycled | 30 K | $920/yr | $4.1 million/yr |
| LH2 | 25 K | $18,300/yr | $29 million/yr |
| LHe | 10 K | $29,300/yr | $330 million/yr |
| LHe, recycled | 5 K | $920/yr | $4.1 million/yr |
Next, the cost of cooling solutions for slow sensors. The most substantial improvement comes from switching from burning LHe to recycling LHe. Switching to other coolants doesn't change the costs much. However, these estimates don't account for a large number of factors; more detailed analysis is needed.
| Ongoing Cost | Conditions | Optimistic | Pessimistic |
|---|---|---|---|
| Hyper-Resolution qPlus AFM | LNe, recycled | $26,100/yr | $14 million/yr |
| Hyper-Resolution qPlus AFM | LH2 | $519,000/yr | $96 million/yr |
| Hyper-Resolution qPlus AFM | LHe, not recycled | $830,000/yr | $1.1 billion/yr |
| Hyper-Resolution qPlus AFM | LHe, recycled | $26,000/yr | $14 million/yr |
| Piezo Creep Bottleneck | LHe, not recycled | $17 million/yr | $1.1 billion/yr |
| Piezo Creep Bottleneck | LHe, recycled | $540,000/yr | $14 million/yr |
Next, the person hours cost. This is comparable to the cost of cryogens or on-site refridgeration.
| Ongoing Cost | Optimistic | Pessimistic |
|---|---|---|
| Fast SPM Sensors | $526,000/yr | $66 million/yr |
| Hyper-Resolution qPlus AFM | $526,000/yr | $220 million/yr |
| Piezo Creep Bottleneck | $526,000/yr | $220 million/yr |
Finally, the helium consumption if not recycled.
| Gaseous Helium Consumption | Optimistic | Pessimistic |
|---|---|---|
| Fast SPM Sensors | 168 m3/yr | 4.6 million m3/yr |
| Hyper-Resolution qPlus AFM | 28,600 m3/yr | 15 million m3/yr |
| Piezo Creep Bottleneck | 588,000 m3/yr | 15 million m3/yr |
All possibilities satisfy the criterion, of being under 30% of US GDP ($8.3 trillion/yr). The worst cases also fall below the annual helium selloff rate (79 million m3/yr). The cases requiring 15 million m3/yr would deplete all US geologic natural gas reservoirs in ~570 years.
Examine the costs with CBN's current approach. I reduced the pessimistic bound by 7.5x, regarding metrics proportional to time delay. There was a non-multiplicative relationship between "feedback loop latency" and "operations/reaction". The upper bound for "operations/reaction" was 600, but the lower bound actually came from CBN. It is fair to combine 60 s/operation with 80 operations/reaction. The 7.5x reduction does not apply to the table after this one, which describes qPlus AFM.
In the worst case, it would take 1300 years and deplete all known US helium deposits, 2.4 times over. In the best case, it would take 30 years while constituting 1% of the US helium market. The sum of all financial costs falls between $540 million and $1.7 trillion. Financially, it is completely doable with the world's resources. Scarcity of helium might not even be an issue, in some scenarios. The reason it's not feasible, is simply because the piezos are too slow.
| Metric | Lower (Optimistic) | Upper (Pessimistic) |
|---|---|---|
| Replicator Atom Count | 40,000,000 | 204,000,000 |
| SPMs in Parallel | 3,500 | 20,000 |
| Time Delay | 30 years | 1,300 years |
| Facility Area (2D) | 2,730 m2 | 62,600 m2 |
| Facility Dimension (1D) | 52.2 m | 250 m |
| Upfront Cost | $17 million | $400 million |
| Ongoing Cost (Hardware) | $17 million/yr | $1.1 billion/yr |
| Ongoing Cost (Salaries) | $526,000/yr | $220 million/yr |
| All Costs over Lifespan of Effort | $540 million | $1.7 trillion |
| Gaseous Helium Usage (Yearly) | 588,000 m3 | 15 million m3 |
| Gaseous Helium Usage (Total) | 17.6 million m3 | 20 billion m3 |
Assume piezo creep is fixed. The next bottleneck is reliance on hyper-resolution qPlus AFM, that can clearly resolve carbon-carbon bonds in aromatic hydrocarbons. We set the measurement bandwidth at 5 Hz, where the frequency shift is 1 Hz. (128 pixels)^2 / 5 Hz = 1 hour. That reproduces the imaging times reported for this technique.
The time required would reduce by a significant amount. The pessimistic case is now within our lifetimes. The greatest possible amount of helium that might be consumed, is 24% of the remaining US stockpile. The sum of all financial costs falls between $3.5 million and $40 billion.
| Metric | Lower (Optimistic) | Upper (Pessimistic) |
|---|---|---|
| Replicator Atom Count | 40,000,000 | 204,000,000 |
| SPMs in Parallel | 170 | 20,000 |
| Time Delay | 2 years | 30 years |
| Facility Area (2D) | 133 m2 | 62,600 m2 |
| Facility Dimension (1D) | 11.5 m | 250 m |
| Upfront Cost | $842,000 | $400 million |
| Ongoing Cost (Hardware) | $830,000/yr | $1.1 billion/yr |
| Ongoing Cost (Salaries) | $526,000/yr | $220 million/yr |
| All Costs over Lifespan of Effort | $3.5 million | $40 billion |
| Gaseous Helium Usage (Yearly) | 28,600 m3 | 15 million m3 |
| Gaseous Helium Usage (Total) | 57,200 m3 | 450 million m3 |
One more optimization is to drastically cut the cost, without changing the time delay. Switch from burning LHe to recycling the LHe. Or swapping it out for LNe, for a marginal reduction in upfront cost and power consumption. The sum of all financial costs falls between $2.8 million and $7.8 billion.
| Metric | Lower (Optimistic) | Upper (Pessimistic) |
|---|---|---|
| Upfront Cost | $1.7 million | $750 million |
| Ongoing Cost (Hardware) | $26,000/yr | $14 million/yr |
| Ongoing Cost (Salaries) | $526,000/yr | $220 million/yr |
| All Costs over Lifespan of Effort | $2.8 million | $7.8 billion |
| Gaseous Helium Usage (Yearly) | 0 m3 | 0 m3 |
| Gaseous Helium Usage (Total) | 0 m3 | 0 m3 |
To shorten the time delay (and slice the cost in proportion), we must move on to faster sensors. It is not currently known what the next bottleneck after piezo creep is. Perhaps it's in the coarse X/Y actuation before using each tripod in inverted mode. This cost could be amortized a little, if ~10 tripods fit on a (100 nm)2 area between coarse actuations. It could also be sample exchange, although migrating from coarse XZ to coarse XYZ should eliminate that bottleneck. Perhaps it is the tooltip being presented, being completely random. If it's not the tooltip you want, you have to move on to a different tripod. This might slow things down an order of magnitude, and there would be no way around it.
How many atoms/day/SPM? Was the idea of 50,000 atoms/day accurate? Let's start out with an 8-hour workday, which we'd expect in the early stages. With a small lab and no automation. This is what we'd report in the first research papers. Compare it to CBN's supposed 6–100 atoms/day reactions/day.
| Sensor or Bottleneck | Optimistic | Pessimistic |
|---|---|---|
| STM (5400 Hz) | 1.9 million reactions/day | 260,000 reactions/day |
| STM (1635 Hz) | 589,000 reactions/day | 78,500 reactions/day |
| qPlus AFM (990 Hz) | 356,000 reactions/day | 47,500 reactions/day |
| qPlus AFM (450 Hz) | 162,000 reactions/day | 21,600 reactions/day |
| qPlus AFM (5 Hz) | 1,800 reactions/day | 240 reactions/day |
| Piezo Creep (1 min) | 6 reactions/day | 6 reactions/day |
Constrain the range of reactions/atom, from 1.55–3.96 to 3.00–3.91. This models methylene taking one reaction to place, one reaction to remove the first hydrogen, and one reaction to remove the second. Carbon dimers are 1.55 reactions/atom when building amorphous carbon. The only repeating unit cell buildable with carbon dimers might be carbon nanotubes. In that case, it could be 0.5 reactions/atom. Either way, it is out of scope for a conservative estimate. Carbon dimers are a speculative optimization regarding "coarse building blocks with several atoms". Anything more complex, like spiroligomer feedstocks, is even more speculative.
| Sensor or Bottleneck | Optimistic | Pessimistic |
|---|---|---|
| STM (5400 Hz) | 630,000 atoms/day | 66,500 atoms/day |
| STM (1635 Hz) | 196,000 atoms/day | 20,100 atoms/day |
| qPlus AFM (990 Hz) | 119,000 atoms/day | 12,100 atoms/day |
| qPlus AFM (450 Hz) | 54,000 atoms/day | 5,520 atoms/day |
| qPlus AFM (5 Hz) | 600 atoms/day | 61 atoms/day |
| Piezo Creep (1 min) | 2 atoms/day | 1.5 atoms/day |
| CBN placing mostly carbon dimers | 4–12 atoms/day | 4–12 atoms/day |
Finally, the atoms/day in a full factory workflow, where SPM operation has been automated.
| Sensor or Bottleneck | Optimistic | Pessimistic |
|---|---|---|
| STM (5400 Hz) | 1.9 million atoms/day | 200,000 atoms/day |
| STM (1635 Hz) | 589,000 atoms/day | 60,300 atoms/day |
| qPlus AFM (990 Hz) | 356,000 atoms/day | 36,300 atoms/day |
| qPlus AFM (450 Hz) | 162,000 atoms/day | 16,600 atoms/day |
| qPlus AFM (5 Hz) | 1,800 atoms/day | 183 atoms/day |
| Piezo Creep (1 min) | 6 atoms/day | 4.5 atoms/day |
| CBN placing mostly carbon dimers | 12–36 atoms/day | 12–36 atoms/day |
Here are some possible next steps, whose investment cost is a fraction of the end-goal semiconductor fab. The steps are ranked roughly from most to least accessible.
Experimental demonstration of lithium niobate scanning probe:
- [Outcome] Eliminate the possibility of a piezo creep bottleneck, which pushes the time delay up to 30–10,000 years.
- [Investment] 1 person-year of full-time work from Philip. He is the person with specialized knowledge on LiNbO3.
- This outcome does not close the uncertainty gap in latency per operation. It is not known whether coarse XY actuation will be the dominant bottleneck (switching between different tripods spaced tens of nanometers apart). However, the investigation will reveal useful insights into coarse actuation.
Theoretical demonstration of atomically detailed replicator design:
- [Outcome] Show that a replicator design does not need a full vacuum enclosure. Thus, the atom count may be reduced by 10x.
- [Investment] 6 person-months of full-time work from Philip. He is the person with specialized knowledge on CAD in the million-atom range.
- This outcome does not definitely prove a replicator can be built with low cost. Nonetheless, it provides important insights into replicator design. It is advised to do this before investing large amounts of money. It could be, that even in theory, there are major issues preventing a 3DOF actuator from working at all. Experimental demonstration would proceed regardless of the results, but it would be comforting to know the results beforehand.
Theoretical demonstration of minimal-cost UHV SPM system:
- [Outcome] More accurate estimates of upfront cost (ongoing costs are already easy to estimate)
- [Investment] Unknown number of person hours; a low-cost investigation will reveal this number
- The early stages might sacrifice cost/SPM for the quickness of getting results. Provided, we have assurance that cost can be minimized in later stages. A thorough theoretical investigation, including FreeCAD 3D models, would be comforting before proceeding with even a single costly SPM.
- We should understand the impact of a situation where cryogens are needed.
Experimental demonstration of tripod synthesis
- [Outcome] Reduce a source of delay between setting up a laboratory, and running an initial mechanosynthesis experiment
- [Investment] Unknown number of dollars and person hours; 2 person-months of full-time work may reveal this number
- If this is done too soon, the effort could be wasted, because we made the wrong tripod (wrong surface chemistry/linkers). An initial target has the Ge-substituted adamantane cage framework, with the bromine-capped ethynyl feedstock.
- An alternative, which may be easier to synthesize, but less useful, is ethynyl-adamantane. This specific tripod would still serve some practical use far-term, as the best possible hydrogen abstraction tool. However, voltage pulses may also compensate for the energetic issues with hydrogen abstraction by Ge-CC.
Experimental demonstration of inverted mode mechanosynthesis:
- [Outcome] 300x reduction in uncertainty of nearly all cost metrics. Eliminates the uncertainty in "feedback loop latency", "operations/reaction", "reactions/atom".
- [Investment] 3 years with multiple employees. Cost tabulated below.
- In order for the "Outcome" to take effect, the atom placement error rate must be extremely low. For example, 10-3 or 10-5. Otherwise, we may still have to resort to hyper-resolution qPlus AFM and/or 4 K for acceptable error rates.
- Judge whether the discovered success rate is sufficient for self-replication without error checking. This analysis would pair well with a theoretical, atomically detailed replicator design.
| Contributor | XPS/MBE required | XPS/MBE not required |
|---|---|---|
| Hardware | $53,000–$111,000 | $18,000–$38,000 |
| Synthesizing Tripods (Salary) | unknown | unknown |
| Synthesizing Tripods (Lab Access) | unknown | unknown |
| Synthesizing Tripods (NMR Access) | unknown | unknown |
| Employee Salary at SPM Lab | $605,000–$1,010,000 | $605,000–$1,010,000 |
The following parameters were used for employee salary:
- random guess of 4 employees
- 35-hour work weeks
- 48 weeks/year
- randomly guessed wage of $30–$50/hr
The following section might be restating the points from the rest of the document. I'm not sure how relevant it is.
Work needed on the computational/theoretical side:
- Understand the issue regarding vacuum chamber housing. Housing consumes 90% of the atoms in a structure. Yet, it is important to have atomically precise frames, with seals (also atomically precise) for feedstocks. The position of the actuator must be defined relative to the place where it builds.
- Can you construct a perfect vacuum seal, when the walls must be broken into several parts ~10,000 to ~100,000 atoms large? Will contaminant gases seep through the vdW gaps?
- Understand the issues with nanopositioner accuracy. Although thermal noise (stiffness/compliance) was minimized in Nanosystems, there are also issues from quantization error and nonlinearities.
- Understand how signals would be transduced from the macroscale, into analog (linear) actuations of a 3DOF or 6DOF nanopositioner at the nanoscale. Especially, near-term, where an artificial replicator becomes the most rudimentary replacement for an SPM.
Work needed on the chemistry side:
- We need to know synthetic routes for a handful of tripods. With detailed procedures for reproducing in a lab without NMR capabilities. Ge-substituted adamantane with Ge-H, Ge-CCI, Ge-CH2I. Regular adamantane with C-CCI. Both R-OH and R-SH linkers.
- Understand how silicon feedstocks (Ge-SiH3) would be generated in-situ. I'm almost certain that, if R-OH linkers are present, they will corrupt the Ge-SiH3 functional group in solution. R-SH might not. This needs more elaboration. Silicon feedstocks will most likely be necessary.
- Understand the low-level details, such as reactions/atom and reaction success rate. What is the atom placement speed?
- Understand whether cryogens are really needed. If so, whether 25–30 K can be used instead of 5–10 K.
- Understand the effect of "blurriness" of the SPM sensor on reaction success rate. This is the next step up, after fixing the horrible conflating factor of piezo creep when attempting to know where the atom is.
Work needed on the hardware side:
- Prove piezo creep can be overcome in a functional SPM with 3 fine DOFs and 2 coarse DOFs (all in the X/Y/Z cartesian directions). This is the greatest priority before any other work.
- Get more concrete estimates to low cost for near-term and far-term vacuum chamber systems. Including the whole system cost, including possible XPS, MBE, and gas liquefaction systems.
- Understand what's to do with transferring things between vacuum chambers. Are the doors between them sources of leaks?
- Understand how to operate large numbers of parallel vacuum chambers with minimal human labor.
- Understand the low-level details of stuffing multiple SPMs into a single chamber.
What we cannot predict:
- Whether a diamondoid nanopositioner even works, with fidelity comparable to lithium niobate. Or whether we can use some other tricks, to get low error rates without linear nanopositioning. The ability to replicate comes from matter going digital, and being able to program a digitized 3D lattice. This programmable lattice constitutes diamondoid parts that make up a replicator, with form closure.
- Whether the replicator can even function outside of UHV and/or cryogenic temperatures (conditions needed to minimize error rate from contaminants or thermal motion).
- How many tries it will take! This document uses a range of 11–200.
Do you want to build a CSRM using STM entirely? You can first, for example, build 1 nanomanipulator of 4,000,000 atoms and then throw out all the STM.