NanoTag: Fine-Grained Tagging Across Disciplines

• NanoTag is a family of tagging systems that convert local physical or logical states into decodable identifiers through engineered interactions.
• Implementations range from vibrating graphene quantum dots and photolithographic marking of semiconductor QDs to DNA-origami color coding and ARM MTE memory safety systems.
• Recurring design features include orthogonal decoding dimensions, precise thresholding, and robust protocols to mitigate interference and metadata errors.

“NanoTag” is a label applied in the literature to several technically distinct tagging, marking, coding, and detection systems. In the cited work, it denotes: a vibrating graphene–quantum-dot acousto-optic radar tag for microfluidic and in-body particle tracking (Gulbahar et al., 2017); a cryogenic, single-color photolithographic method for permanently marking individual semiconductor quantum dots (Sawicki et al., 2015); a DNA-origami colorimetric path-tagging scheme for gliding filaments (Micolich, 2019); a time-domain optical fingerprinting platform based on single upconversion nanoparticles (Liao et al., 2020); isotopically engineered single-walled carbon nanotube Raman tags for multiplexed molecular imaging (Liu et al., 2010); and a byte-granular overflow-detection system for ARM Memory Tagging Extension (MTE) (Li et al., 26 Sep 2025). Across these usages, the common theme is not a shared substrate or protocol, but the conversion of local physical or logical state into a decodable identifier.

1. Scope, nomenclature, and technical taxonomy

In the cited literature, “NanoTag” is not a single standardized technology. It refers instead to a family of tagging paradigms that differ in substrate, signal modality, and decoding method.

Usage of “NanoTag”	Substrate	Primary function
CSSTag / optical nano-radar (Gulbahar et al., 2017)	Graphene resonator with CdSe/ZnS QDs	CSS identification and TOF particle tracking
Cryogenic QD marking (Sawicki et al., 2015)	SU-8 mark over CdTe/ZnTe QD	Permanent relocation of individual emitters
DNA color-run NanoTag (Micolich, 2019)	DNA-origami barrel on filaments	Path encoding by accumulated fluorophore tags
t$_2$-Dot NanoTag (Liao et al., 2020)	Upconversion nanoparticles	Multiplexed optical barcoding
SWNT Raman NanoTag (Liu et al., 2010)	Isotopically engineered SWNTs	Five-color molecular imaging
ARM MTE NanoTag (Li et al., 26 Sep 2025)	Tagged heap allocations	Byte-granular overflow detection

The first five usages operate through nanoscale or molecular structure. The sixth uses “NanoTag” as a system name in computer systems research. This suggests that the term has evolved into a general label for fine-grained tagging systems rather than a domain-specific designation.

2. Graphene VFRET NanoTag for nanoscale radar and particle tracking

In "CSSTag: Optical Nanoscale Radar and Particle Tracking for In-Body and Microfluidic Systems with Vibrating Graphene and Resonance Energy Transfer" (Gulbahar et al., 2017), the NanoTag is a signaling based nanoscale acousto-optic radar and microfluidic particle tracking system built around a Vibrating FRET (VFRET) optical modulator. A single VFRET unit consists of a doubly-clamped, single-layer graphene membrane of length $L_g$, width $W_g$ and thickness $h_g$, carrying in-plane pre-tension $F_T$. Along its two free edges a linear chain of CdSe/ZnS donor QDs is drop-cast, and a fixed array of acceptor QDs is held at a separation $d_{AD}$. Ultrasonic excitation drives the membrane at its fundamental resonance, periodically modulating donor–acceptor separation and therefore acceptor emission through Förster resonance energy transfer.

In the small-deflection regime, the membrane is treated as an effective harmonic oscillator. The fundamental resonance is written as

$$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$

In fluid, the resonance down-shifts to

$$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$

and with QD load $\Delta m$ further to

$$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$

The transfer efficiency of a donor–acceptor pair at separation $L_g$0 is

$L_g$1

The defining step is the conversion of mechanical oscillation into an optical chirp spread spectrum waveform. By stacking donor–acceptor layer pairs at calibrated vertical intervals $L_g$2 and $L_g$3, each mechanical half-period causes the donor cluster to sweep past multiple acceptor planes, producing a pulse train with time-varying instantaneous frequency. The resulting CSS chirp is approximated as

$L_g$4

with $L_g$5 and $L_g$6. Distinct tags are programmed by choosing different subsets of acceptor layers, yielding different time–frequency supports $L_g$7.

Detection follows a classical radar formulation. With $L_g$8 tags, the photodetector output is modeled as

$L_g$9

and matched filtering or cross-correlation produces

$W_g$0

whose peak gives $W_g$1. The position estimate along the ultrasonic axis is then

$W_g$2

The details also state that Doppler shift can be estimated from the phase slope at the correlation peak, yielding velocity information.

The reported simulation regime is explicit. Monte-Carlo trials track $W_g$3 particles in water with $W_g$4 m/s, $W_g$5m, $W_g$6 nm, $W_g$7 kHz, and light excitation of $W_g$8 W/m$W_g$9/nm (Gulbahar et al., 2017). For single-particle tracking, the average delay error falls below one time-bin $h_g$0 ns$h_g$1 once $h_g$2 dB for Tag 1 or $h_g$3 dB for Tag 2. In multiple-particle tracking, zero-error delay estimation is reached above $h_g$4 dB SNR when the inter-particle TDOA avoids the worst-case correlation nulls. Position resolution approaches $h_g$5m, and update rates up to $h_g$6 kHz are demonstrated. The modulator dimension is $h_g$7m $h_g$8 $h_g$9m $F_T$0 $F_T$1m with several picograms of weight. The paper places the platform in microfluidic cytometry, in-body single-cell tracking, and nano-biological diagnostics, and notes fabrication issues including precision patterning of graphene membranes, controlled QD drop-casting, and reliable membrane clamping.

3. Cryogenic NanoTagging of individual semiconductor quantum dots

In "Single-color, in situ photolithography marking of individual CdTe/ZnTe Quantum Dots containing a single Mn$F_T$2 ion" (Sawicki et al., 2015), NanoTagging denotes a single-color method for permanent marking of the position of individual self-assembled semiconductor quantum dots at cryogenic temperatures. The sample consists of self-assembled CdTe QDs with density $F_T$3 cm$F_T$4, grown by MBE in a ZnTe barrier on GaAs, with the QD plane 100 nm below the surface. After solvent cleaning, negative SU-8 2002 resist is spin-coated at 6,100 rpm to $F_T$5m thickness and soft-baked at $F_T$6C for 1 min.

Position determination is performed by low-temperature micro-photoluminescence mapping at $F_T$7 K in a He-flow cryostat. Excitation is above the ZnTe barrier edge at $F_T$8 eV $F_T$9 nm$d_{AD}$0 with a cw diode laser. A $d_{AD}$1, NA$d_{AD}$2 objective with 6.5 mm working distance forms a $d_{AD}$3m spot, mounted on piezo-stages with 20 nm positioning resolution. A few-$d_{AD}$4m$d_{AD}$5 area is mapped in a 20 nm step-scan, typically with $d_{AD}$6–0.1 kW/cm$d_{AD}$7 and $d_{AD}$8 s per point, yielding $d_{AD}$9 nm uncertainty in the QD centre. Because SU-8 sensitivity at 405 nm is low, this mapping dose remains below the exposure threshold.

Permanent marking uses the same beam. Once a target QD is centered, the power is ramped to $$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$0 kW/cm$$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$1 for $$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$2 s, exposing a tiny circular SU-8 region directly above the emitter; a $$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$3m$$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$4 landmark can also be written (Sawicki et al., 2015). Post-exposure bake at $$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$5C for 1 min and development for 1 min produce a permanent micropatch of SU-8 over the selected QD. The optical model uses a Gaussian beam

$$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$6

with $$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$7m, and a diffraction-limited lateral resolution

$$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$8

For $$f_0 \;=\; \frac{1}{2\pi}\,\sqrt{\frac{k_\mathrm{eff}}{m_\mathrm{eff}}} \;\approx\; \frac{1}{2\,L_g}\,\sqrt{\frac{F_T}{\rho_g\,W_g}}.$$9 nm and NA$$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$0, $$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$1 nm, although the smallest exposed SU-8 spot reported in practice is $$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$2m.

The quantitative performance is specific: spatial resolution is a $$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$3m diameter SU-8 spot after development; mapping precision is $$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$4 nm; marker-to-QD alignment is $$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$5m; and tagging yield is 16/16 QDs found under their respective marks, corresponding to $$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$6 yield (Sawicki et al., 2015). The low-temperature environment preserves spectral resolution down to $$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$7eV, and SU-8 remains crack-free if cooled slowly, with $$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$8 h from 300 K to 10 K.

The method is demonstrated on a single CdTe/ZnTe QD containing a single Mn$$f_w \;=\; \frac{f_r}{\sqrt{1 + \Gamma_w \,\rho_w\,L_g/(\rho_g\,h_g)}},$$9 ion. At zero field, the neutral exciton splits into six lines at 2034 meV due to $\Delta m$0–Mn exchange. In Faraday configuration up to 10 T, $\Delta m$1/$\Delta m$2 polarization-resolved PL tracks linear Zeeman and quadratic diamagnetic shifts, with anticrossings around $\Delta m$3 T in $\Delta m$4. In Voigt configuration up to 10 T, sixfold splitting persists with field-induced Mn spin quantization. The Hamiltonian is given as

$\Delta m$5

with

$\Delta m$6

$\Delta m$7

$\Delta m$8

Best-fit parameters include $\Delta m$9, $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$0, $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$1 meV/T$$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$2, $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$3 meV, $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$4 meV, $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$5 meV, $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$6 meV, $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$7 meV, and $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$8 K. The paper states that the excellent match confirms that the polymer marker does not perturb the QD’s magneto-optical response.

4. DNA-origami NanoTag for colorimetric path encoding

In "Colorimetric path tagging of filaments using DNA-based metafluorophores" (Micolich, 2019), NanoTag is a nanoscale extension of a sporting event called a “color run.” The object being tagged is a filament, specifically a gliding microtubule, wrapped with a DNA origami barrel that acts as a nanoscale “t-shirt.” The basic scaffold is a 60 nm $$f_0 \;=\; f_w \,\bigl(1 - \tfrac{\lvert\Delta m\rvert}{2\,m_d}\bigr).$$9 90 nm rectangular DNA origami tile that is rolled into a cylinder with circumference $L_g$00 nm and length $L_g$01 nm. On the inner surface are up to 2,000 uniquely addressable ss-DNA handles spaced $L_g$02 nm apart, used for anti-handle–antibody attachment to microtubules; on the outer surface, each handle can host a single fluorophore-bearing tag.

The optical encoding is based on DNA-based metafluorophores. The top face of the tile is divided into three parallel stripes of 44 handles each, assigned to red, green, and blue. Because quenching and FRET are minimized by spatially separating the stripes, the total emission is approximated as the linear sum

$L_g$03

with corresponding color vector

$L_g$04

Using integer levels $L_g$05 for each color yields $L_g$06 distinct non-zero codes; requiring each $L_g$07 yields $L_g$08 codes with $L_g$09 read accuracy (Micolich, 2019).

Tag acquisition is station-specific. Each barrel carries outer ss-DNA handles $L_g$10 specific to station $L_g$11, and complementary anti-handle tags $L_g$12 hybridize through

$L_g$13

The summary gives typical $L_g$14–$L_g$15 M$L_g$16s$L_g$17, $L_g$18–$L_g$19 s$L_g$20 for $L_g$21 bp duplexes, and low-nM $L_g$22. A toehold-mediated strand-displacement countermeasure is proposed to remove stray tags: $L_g$23 Typical strand-displacement rates are $L_g$24–$L_g$25 M$L_g$26s$L_g$27 for $L_g$28-nt toeholds.

The color-stations are implemented with a two-layer PDMS microfluidic top-plane bonded over the network flow-cell, with a 10–20 $L_g$29m-thick polycarbonate dialysis membrane whose $L_g$30 nm pores allow anti-handle–fluorophore tags to cross but block microtubules. Each station is a local chamber of approximately $L_g$31m $L_g$32m cross-section, perfused at 0.1–1 $L_g$33L/min with a 10–100 nM tag solution. Network flow is driven at $L_g$34L/min of ATP-containing buffer (Micolich, 2019).

Path encoding is additive. A gliding microtubule carries $L_g$35 barrels per 5 $L_g$36m of filament. If station $L_g$37 contributes $L_g$38 dyes of color $L_g$39, then after traversing a sequence $L_g$40,

$L_g$41

The readout is lenseless on-chip fluorescence using a CMOS backplane. Read-out sites are hundreds of $L_g$42m downstream, with custom backplane tiles comprising 3–9 sub-pixels with RGB filters. Exposure times are $L_g$43–100 ms, and the raw signal is modeled as

$L_g$44

with shot noise $L_g$45 and read noise $L_g$46. The triplet $L_g$47 is decoded by nearest-neighbor classification in color space.

The performance figures are framed as capacity and error control. A single network can process $L_g$48 filaments in parallel; at gliding speeds of 1 $L_g$49m/s and read-out every 100 $L_g$50m, throughput exceeds $L_g$51 read-events per hour. The primary error mode is false positives from stray tags, with

$L_g$52

With the strand-displacement sink, $L_g$53 is reduced to $L_g$54 per station, and across $L_g$55 stations the overall path-decoding error is approximated as $L_g$56 (Micolich, 2019).

5. Optical NanoTags: time-domain fingerprints and Raman codes

A major use of “NanoTag” is nanoscale optical coding. Two distinct implementations illustrate this: time-domain emissive fingerprints from upconversion nanoparticles and isotopically shifted Raman signatures from single-walled carbon nanotubes.

In "Wide-field Decodable Orthogonal Fingerprints of Single Nanoparticles Unlock Multiplexed Digital Assays" (Liao et al., 2020), the NanoTag or t$L_g$57-Dot platform turns single upconversion nanoparticles into massively multiplexable barcodes by engineering their full time-domain emissive profile. The key observable is the time trace $L_g$58 following a short excitation pulse. With $L_g$59 denoting the excited-state population,

$L_g$60

which yields

$L_g$61

or equivalently

$L_g$62

The extracted parameters are $L_g$63, $L_g$64, and $L_g$65, with

$L_g$66

Experimentally, a 200 $L_g$67s excitation pulse is used and $L_g$68 is recorded in 75 consecutive 50 $L_g$69s gated frames with an intensified sCMOS camera.

The t$L_g$70 profile is tuned by interfacial energy migration shells, dopant concentration, core/shell thickness, and surface-quencher isolation. The reported range spans $L_g$71–1000 $L_g$72s and $L_g$73–2000 $L_g$74s, with 42 distinct UCNP batches and within-batch variation $L_g$75 CV (Liao et al., 2020). Coding capacity scales as

$L_g$76

With 2 excitation wavelengths, 3 spectral channels, and 42 t$L_g$77 profiles, the demonstrated total is $L_g$78 distinct codes. The optical readout uses a wide-field microscope with intensified sCMOS in Integrate-On-Chip mode at 250 Hz; 75-point normalized traces are decoded by a convolutional neural network with two 1D-convolutional layers, two fully connected layers, softmax output, Adam optimizer with learning rate 0.005, batch size 256, and 50 epochs. Reported per-batch single-particle classification accuracy is 91–100% with mean $L_g$79, and throughput is $L_g$80 codes/s. Demonstrations include sub-diffraction data storage, security inks, a five-plex single-molecule digital assay for HBV, HCV, HIV, HPV-16, and Ebola with $L_g$81 specificity, and upconversion SIM at 184.8 nm resolution.

In "Multiplexed five-color molecular imaging of cancer cells and tumor tissues with carbon nanotube Raman tags in the near-infrared" (Liu et al., 2010), NanoTags are single-walled carbon nanotubes synthesized with five different C$L_g$82/C$L_g$83 isotope compositions and well-separated Raman peaks. The five G-band positions are 1529, 1546, 1559, 1575, and 1590 cm$L_g$84, with adjacent colors separated by $L_g$85 cm$L_g$86. After purification, dispersion, and PEGylation, the nanotubes are conjugated to targeting ligands including Erbitux, Rituxan, Herceptin, goat anti-CEA IgG, and c(RGDyK)-SH. Raman intensity follows

$L_g$87

The imaging configuration is a 785 nm diode laser at 80 mW with a 50$L_g$88 objective, $L_g$89m spot, and 1 mm confocal pinhole. At 300 nM SWNT and OD$L_g$90, a 1 $L_g$91m $L_g$92 1 $L_g$93m pixel with 0.5 s integration yields 1,000–5,000 counts in the G-band, while background noise is $L_g$94–40 counts, giving S/N $L_g$95 (Liu et al., 2010). Cells are stained with 10 nM of each color for 1 h at $L_g$96C, and five-color maps are obtained by deconvolution against the library of pure-SWNT G-bands. The stated detection limit is $L_g$97 of single-cell saturation signal; non-specific binding is $L_g$98 of positive signal; and semi-quantitative receptor profiling agrees with flow cytometry within 10%.

The ex vivo tumor demonstration uses LS174T xenografts sectioned to 5 $L_g$99m and incubated with a five-plex NanoTag mix. The maps resolve CEA, EGFR, integrin $W_g$00, CD20, and Her2. A specific biological finding is that EGFR on LS174T cells is near background in vitro $W_g$01 of CEA signal$W_g$02 but reaches $W_g$03 of CEA intensity ex vivo, corresponding to an $W_g$04-fold increase in average signal in vivo versus in vitro $W_g$05 (Liu et al., 2010). In this Raman implementation, the distinguishing features are sub-2 nm effective spectral width, negligible autofluorescence background, and absence of photobleaching under repeated scans.

6. NanoTag as byte-granular overflow detection on ARM MTE

In "NanoTag: Systems Support for Efficient Byte-Granular Overflow Detection on ARM MTE" (Li et al., 26 Sep 2025), NanoTag is a memory-safety system rather than a material or optical tag. ARM MTE attaches a 4-bit memory tag to each 16-byte tag-granule and a matching 4-bit address tag to pointers. Hardware detects mismatches on loads and stores, but the smallest unit of checking is 16 bytes. As stated in the paper, an off-by-$W_g$06 overflow that stays within one 16-byte granule is an intra-granule overflow and can go undetected.

NanoTag addresses this by identifying short granules, sampling some of them, and turning them into tripwires. For an allocation of size $W_g$07, a short granule exists when $W_g$08, with only $W_g$09 valid bytes in the final granule. NanoTag sets the memory tag of that granule to the special tripwire tag, stores the object’s real pointer tag in the low 4 bits of the last byte of padding, and uses the remaining padding for an AccessCount counter (Li et al., 26 Sep 2025). Any legitimate access to the valid bytes of the tripwire granule then causes a hardware tag mismatch, which is routed to a software handler.

The byte-granular check in the fault handler is defined by:

1. rejecting if the memory tag or address tag is zero;
2. rejecting if the address tag does not match the stored metadata tag;
3. computing

$W_g$10

and treating the access as safe iff $W_g$11. Benign accesses are recovered by a three-step mechanism named Delegation–Escalation–Revocation: temporarily swapping the tripwire tag with the stored real tag so the instruction can complete, overwriting the next instruction with a one-byte BRK trap, and then restoring the original tagging state and incrementing AccessCount when the trap fires. If AccessCount exceeds a threshold, the tripwire is removed from that hot granule.

Integration is implemented in the Scudo Hardened Allocator, the default allocator on Android since Android 11. NanoTag modifies Scudo’s primary allocator only; it does not change the secondary allocator or user ABI (Li et al., 26 Sep 2025). Evaluation is reported on a Google Pixel 8 Pro running Android 15 with Termux/Ubuntu 22.04 jail, compiled with clang14 and -O3. On Juliet v1.3 CWE122 heap-buffer-overflow, ASAN detects 98.66%, Scudo+MTE in SYNC or ASYNC detects 75.6%, and NanoTag with default parameters detects 97.57%. For CWE415 and CWE416, NanoTag is approximately Scudo’s 96–98%. On SPEC CPU 2017 SPECrate Integer, with baseline defined as Scudo with MTE disabled, overheads are 4.0% for Scudo+MTE(ASYNC), 11.98% for Scudo+MTE(SYNC), 12.50% for NanoTag, and 95.11% for ASAN. On Geekbench 6, the reported overheads are +1.96% for Scudo+MTE(ASYNC), +3.76% for Scudo+MTE(SYNC), +4.99% for NanoTag, and +1348.6% for Valgrind-Memcheck. In AFL++ fuzzing on Magma 1.2.1 targets, NanoTag slows fuzzing by geomean 15.86%, compared with 111.20% for ASAN.

This usage broadens the meaning of NanoTag beyond nanomaterials or nanoscale optics. The commonality with the physically nanoscale systems is architectural rather than material: a latent state is encoded in a compact tag and decoded only when a readout event occurs.

7. Cross-domain interpretation and recurring design patterns

Despite the heterogeneity of implementations, the cited NanoTag systems share several recurring technical motifs. First, each system couples a local state to a decodable signature. In the VFRET radar tag, membrane motion modulates FRET and yields a CSS waveform (Gulbahar et al., 2017). In cryogenic QD marking, localized SU-8 cross-linking converts the optical position of a single QD into a permanent micropatch (Sawicki et al., 2015). In the DNA barrel system, path history is transformed into an RGB stoichiometric vector (Micolich, 2019). In the UCNP platform, excitation-history dynamics are encoded in $W_g$12, $W_g$13, and $W_g$14 (Liao et al., 2020). In the SWNT system, isotope composition generates spectrally orthogonal Raman identifiers (Liu et al., 2010). In ARM MTE NanoTag, allocator state is converted into hardware-visible tag metadata (Li et al., 26 Sep 2025).

Second, decoding generally relies on orthogonality in one or more dimensions. The CSSTag work uses distinct time–frequency supports and matched filtering. The DNA color-run system uses orthogonal handle sets and nearest-neighbor classification in color space. The t$W_g$15-Dot platform multiplies excitation wavelength, emission color, and time-domain fingerprint. The SWNT Raman system exploits five separated G-band peaks under a single 785 nm laser. The software NanoTag uses mismatch between address tag and memory tag as a trigger for selective software analysis.

Third, the dominant limitations are those of interference, thresholding, and metadata integrity. The graphene radar tag reports mutual chirp interference in multiple-particle tracking and notes the need for careful correlation analysis (Gulbahar et al., 2017). The QD marking technique is practically limited to $W_g$16m marks by objective NA and resist threshold (Sawicki et al., 2015). The DNA path-tagging scheme identifies stray-tag contamination as the primary error mode and proposes a strand-displacement sink (Micolich, 2019). The UCNP platform notes overlapping codes at low signal levels and dependence on customized optics and machine-learning pipelines (Liao et al., 2020). The MTE-based NanoTag retains the 6.25% chance of random tag collision inherent to 4-bit tags and can miss cases when the faulting instruction is the last in a function (Li et al., 26 Sep 2025).

Taken together, these systems show that “NanoTag” has become a cross-disciplinary name for fine-grained tagging architectures whose readout may be optical, spectroscopic, mechanical, chemical, or architectural. A plausible implication is that the term is best understood functionally: a NanoTag is a compact tag whose identity emerges from engineered interactions and a corresponding decoding pipeline, rather than from any single material platform.