Fabrication · Manufacturing Research

The Three Hard Problems

Between simulation-confirmed device physics and a working chip stand three fabrication problems that no current process solves simultaneously. Each section below maps the problem, the state of the field, and the available solutions.

Blog Post CVD & Chirality Notes Patents
Problem 1
Priority approach filed

Chirality Yield

CVD grows a chirality distribution. Only (8,8) armchair works. Addressed by Thatte1 metallocene CVD process (2006 priority).

Thatte1
Problem 2
Hybrid path

Optical Addressing at Scale

Diffraction (~355 nm) prevents per-device photon delivery at chip density. Reframed: separating addressing (electrical) from gating (optical).

Thatte2
Problem 3
SNR headroom

Disorder Robustness

Gap variation, defect density, and axial misregistration all degrade trit margin. 54 dB SNR provides substantial tolerance; symmetry requires calibration.

Fabrication Thatte1
Problem 1

Chirality Yield

The Problem

The THATTE device inner conductor must be a metallic (8,8) armchair SWCNT. The (8,8) designation is not approximate: other armchair tubes ((7,7), (9,9)) have different diameters, different van der Waals coupling constants, and different conductance quantum values. Wrong chirality means wrong coupling — either too weak to gate or too strong to modulate.

Catalytic CVD nucleation is stochastic. Catalyst engineering narrows the distribution toward a family rather than a specific (n,m) pair. High single-chirality yield (>90%) has been demonstrated only at small scale (tens to hundreds of tubes) under highly controlled conditions.

MethodYieldScaleBarrier
Unoptimised CVD ~2–5% Bulk No selectivity
CoMo/Fe-Co CVD ~15–30% Bulk Family only
Template / cloning >90% Tens of tubes Not scalable
Gel chromatography >99% mg in solution Destroys coaxial
Open sub-problems
  • In-situ Raman verification of (8,8) at wafer scale
  • Yield metric for coaxially encapsulated (8,8) vs. isolated (8,8)
  • Growth time vs. yield trade-off in metallocene-seeded CVD
Solution
Thatte1 — Primary approach
Metallocene CVD: in-situ chirality-selective encapsulation

A unified metallocene CVD process grows (8,8)-enriched SWCNT directly inside an MWCNT channel in a single step, combining chirality selection with coaxial encapsulation. The metallocene precursor's molecular geometry biases nucleation toward armchair configurations while the MWCNT channel constrains the diameter range.

The specific catalyst formulation, temperature profile, and carbon source chemistry are the filed claims of Thatte1. The 2006 priority date predates published literature on in-situ encapsulation for armchair-selective growth.

Why post-growth sorting is not the answer
Sorting destroys coaxial geometry

Gel chromatography and density-gradient ultracentrifugation achieve >99% single-chirality purity, but both require disaggregation of the nanotube bundle into solution. This destroys the SWCNT@MWCNT coaxial geometry that is the whole point of the device. In-situ growth is the only route that achieves both chirality selectivity and coaxial encapsulation simultaneously.

Problem 2

Optical Addressing at Scale

The Problem

The THATTE device is gated by a ~693 nm photon absorbed by the MWCNT outer wall. Photon absorption is the mechanism: it detunes MWCNT energy levels from the SWCNT, weakening inter-wall quantum coupling and raising conductance from 1.47 to 1.91 × G0. Without scalable photon delivery, there is no chip.

Two independent walls make free-space optical addressing infeasible at chip density:

  • Spatial: Abbe limit at 693 nm is ~347 nm. Device pitch at 1010 density is 10–50 nm. Free-space optics cannot address a single device.
  • Temporal: THATTE trit rate is 500 GHz–10 THz. Free-space modulators cap below 100 GHz commercially.
ApproachConfinementBWCNT status
Si photonic wire ~250 nm ~100 GHz Single device
SiN waveguide ~400 nm ~50 GHz Not published
Plasmonic slot <50 nm >1 THz Research only
PhC nanobeam ~100 nm ~10 GHz High Q, narrow BW
Open sub-problems
  • Evanescent coupling efficiency: SiN waveguide to MWCNT at 693 nm
  • Thermal crosstalk between adjacent waveguides and CNT devices
  • EO modulator bandwidth at 500 GHz–1 THz in CMOS platforms
  • WDM crosstalk at <1 nm channel spacing at chip scale
Solutions
Option A — Primary · Already filed
Electrical addressing + optical gate enable

Keep electrical contacts. Each device is individually addressable by its own electrical line — exactly what the semiconductor industry has mastered. Light provides a uniform gate-enable across the array. The per-device electrical line carries the logic; the photon is an enable signal, not a data carrier.

The diffraction problem disappears: light no longer needs to resolve individual devices. Density scales with electrical routing, not with optics. This is the primary embodiment in Thatte1 — the scalable path, and the honest answer to “but how does it scale?”

What it costs: The contactless story is gone. The photon becomes a clock, not a logic input. The 12-state WDM richness at the single-device level is sacrificed in favour of a fast ternary current-mode array. Smaller dream — but a real one, buildable this decade.

Option B — Photonic co-processor
Per-device waveguide + evanescent coupling

Each device sits on a SiN waveguide that carries its photon directly, coupling evanescently into the MWCNT wall. Per-device addressing comes for free: each waveguide is a private channel. Speed is excellent.

The catch: a SiN waveguide is ~450 nm wide, and neighbours must be several microns apart to prevent leakage. The device is 2 nm. Density crashes to millions of elements, not billions. Not the CPU — but a valuable fast photonic co-processor or interconnect fabric.

Option C — WDM cluster multiplier
One optical bus, each device tuned to its wavelength

One shared waveguide carries many wavelengths; each device is tuned by its length to respond only to λN. Serves ~100 devices per bus (optical bandwidth + cavity Q limits). Tops out at hundreds of devices per bus, not millions.

Best used as a hybrid add-on to Option A: WDM bus feeding ~100-device clusters, electrical routing between clusters. Multiplies Option A rather than replacing it. Patent Thatte2 covers this architecture.

The honest frontier — Unsolved by anyone
Near-field, plasmonic, and holographic schemes

Near-field/plasmonic tricks squeeze light to 10–20 nm — still 10× too coarse for 2 nm device pitch. Holograms and spatial light modulators offer parallel illumination but remain diffraction-limited to ~350 nm spots. There is no known way to deliver an individually-chosen photon to each of a billion 2 nm devices at 2 nm spacing. This is not a failure of this project — it is a wall in physics as currently understood.

The Reframe

Electrons are born for dense individual addressing. Photons are born for fast contactless gating. Every version of this architecture that actually scales at chip density is a hybrid, and the simplest hybrid is Option A. The device was never meant to be an all-optical computer — the optical nature is a gift for speed and gate physics, not for routing. Let each medium do its own work.

Problem 3

Disorder Robustness

The Problem

Real fabrication produces a statistical ensemble. Every device differs from its neighbour in inter-wall gap, defect density, and coaxial alignment. The question is whether the device physics tolerates this spread.

SourceRangeEffect on trit margin
Gap variation ±0.03–0.05 nm 5–20% ΔGmod
Defect density 1 per 100–500 nm G: 1.91 → toward 1.47×G0
Axial offset 0.05–0.2 nm Trit ±1 asymmetry 3–5%

Axial misregistration is the sharpest concern: off-axis coupling breaks I(+V) = −I(−V) symmetry and introduces a trit offset that the circuit layer must compensate per device.

Open sub-problems
  • NEGF simulation with gap disorder and defect disorder simultaneously
  • Statistical trit margin vs. process sigma model for a real CVD population
  • Circuit-level calibration for per-device trit offset at chip scale
  • Defect reduction by post-growth annealing without chirality redistribution
Solution
The SNR headroom argument
54 dB margin absorbs most disorder

The NEGF simulation result of SNR > 2000 (54 dB) is the central argument for disorder robustness. The trit margin is not set by a small voltage difference (as in DC CMOS) but by a photon absorption event with a near-binary outcome: absorbed or not. Even if defects halve the conductance modulation depth — from 0.44 G0 to 0.22 G0 — SNR remains above 1000, adequate for digital trit discrimination.

Gap variation changes coupling by ±10% for a ±0.05 nm departure from nominal (exp(−αΔd), α ≈ 2.1 nm−1). This is well within trit margin.

Fabrication side — Thatte1
In-situ encapsulation minimises misregistration

Growing the SWCNT inside the MWCNT in a single CVD step produces near-coaxial geometry from the start, without the axial offset risk of post-growth insertion. The Thatte1 fabrication process addresses misregistration at source rather than correcting it in the circuit.

Circuit side — Thatte2
Differential readout + per-device calibration

Residual trit asymmetry (3–5% for 0.1 nm lateral offset) is detectable by the readout circuit. The Thatte2 gate library includes differential readout and a trit calibration offset register per device to compensate. This adds a one-time calibration step to the manufacturing flow but does not require process perfection.

Why They Cannot Be Solved Independently

The Coupling Problem

Each problem has partial solutions in the literature. They remain hard because the solutions conflict:

Solution for…Conflicts with…Mechanism
Chirality selectivity (slow, low-T CVD) Disorder (defects) Slow growth gives more time for structural rearrangement at the growth front → higher defect density
Defect reduction (high-T anneal) Chirality yield High-temperature annealing reshuffles tube chirality distribution
Post-growth chirality sorting Disorder (coaxial geometry) Sorting requires disaggregation; destroys SWCNT@MWCNT encapsulation
Planar substrate for photonic waveguide integration Disorder (misregistration) Planar substrate constrains MWCNT orientation and insertion geometry
A unified fabrication process that solves all three constraints simultaneously is worth more — in engineering and patent terms — than three separate technique patents, because the coupling means the solutions must co-evolve. Patent Thatte1 is structured accordingly.
Patent Coverage

What the THATTE Stack Addresses

ProblemSub-problemPatentCoverage
Chirality yield In-situ (8,8) growth during encapsulation Thatte1 Catalyst formulation and process conditions (filed, not disclosed)
Coaxial geometry preservation Thatte1 Unified single-step CVD (2006 priority date)
Optical addressing Electrical crossbar + global optical gate (primary) Thatte1 Primary embodiment; palladium contacts + AC bus + flood illumination
Per-device waveguide / evanescent coupling Thatte2 Photonic co-processor embodiment (filed)
WDM multi-wavelength gate architecture Thatte2 Dual-photon mode, WDM-compatible gate library (filed)
Disorder robustness In-situ encapsulation for coaxial geometry Thatte1 Minimises axial misregistration at source
Differential readout + trit calibration Thatte2 Per-device trit offset correction in gate library

International filings (PCT, USPTO, EPO, IPOS) are targeted for March 2027. Specific process parameters and circuit implementations are not disclosed publicly until that filing is secured. The coverage above reflects the scope filed with the Indian Patent Office (April 2026).

Related Reading
Blog Post · 12 Jul 2026

The Three Hard Problems

The narrative introduction: why the three problems are coupled and why they gate the simulation-to-chip transition.

Blog Post · 30 Mar 2026

CVD and Chirality Control

The CVD science, the chirality problem, and the 2006 priority date for the fabrication patent.

Blog Post · 17 Apr 2026

NEGF Quantum Transport

The simulation that confirmed the device: ±74 µA trit currents, SNR > 2000, and the 54 dB headroom that disorder robustness depends on.

Stay Updated

Subscribe to maniTLab

New research, patent updates, and essays — direct to your inbox. No spam, no tracking.

Get in Touch

Write to Manish

Research collaboration, licensing enquiries, technical questions, or press requests.

Replies to manish@manitlab.org