Self‐Aligned Contact Doping for Performance Enhancement of Low‐Leakage Carbon Nanotube Field Effect Transistors

Carbon nanotube (CNT) field effect transistors (CNFETs) show promise for the next generation VLSI systems due to their excellent scalability, energy efficiency, and speed. However, high leakage current is a drawback of large diameter CNTs (diameter (DCNT) ≥ 1.4 nm) due to the small electronic band gap (EG) ≤ 0.6 eV and effective mass. This work investigates the on‐current and off‐current tradeoff for two populations of semiconducting‐enriched CNT with DCNT ≈ 1.0 nm displaying a simultaneous 50x improvement in minimun current (IMIN) with 2.5x degradation in contact resistance compared to DCNT ≈ 1.4 nm using a Pd side‐bonded contact. A method to enhance the performance of low‐leakage CNFETs is demonstrated using sub‐monolayer self‐aligned contact doping with 0.8 nm of MoOX, which delivers a 57% reduction in contact resistance to DCNT ≈ 1.0 nm. Robustness is verified after annealing at 200 °C for 30 min and monitoring stability across 6 months post‐fabrication with no change in electrical behaviors.


Introduction
Semiconducting carbon nanotubes (CNT) are a promising candidate channel material for future highly scaled and high-performance CMOS transistors.Conventional bulk semiconductor materials (e.g., Si and Ge) suffer significant mobility degradation when etched into ultra-thin FinFET or nanosheet channels to preserve electrostatic control as the physical gate length of modern transistors shrinks below 20 nm. [1]Lowdimensional channels such as CNTs maintain excellent transport properties due to an innate ultra-thin body ≈ 1.0 nm, with an electronic band gap modulated by diameter to enable unique engineering of efficiency and speed. [1]Recently, numerous device-level performance metrics have been reported, including excellent subthreshold slope (65 mV dec −1 ) at 15 nm gate length, [2] high driving current with dense CNT array (>2 mA μm −1 at V DS of −0.8 V), [3] and low contact resistance (6.5 kΩ per CNT) with 10 nm contact length. [4][4][5][6] Lin et al. [7] experimentally measured the minimun current (I MIN ) versus E G trend for V DS of −0.5 V in CNT MOSFETs due to band-to-band tunneling and concluded E G < 0.65 eV (D CNT > 1.3 nm) exceeds the upper-bound 100 nA/μm off-current density for highperformance operation.Increasing CNT E G (decreasing D CNT ) is a clear strategy to target a low power consumption regime with 1.0 nm suggested by S.K. Su et al. to trade off drive current and leakage [1] optimally.Methods to enhance the performance of lowleakage carbon nanotube field effect transistors (CNFETs) have not been investigated by experiments to date.
Several alternative methods to suppress the leakage current of CNFETs (besides increasing E G ) have been reported, such as using a Dirac source FET with graphene contact to CNT, [8] additional gates to electrostatically dope the drain extension to suppress band-to-band tunneling, [9] or optimizing spacer to suppress ambipolar tunneling from drain contact. [10,11]However, these methods either have significant area penalties due to extra gates or do not address the leakage current challenges in small band gap CNTs. [7]n this work, a back-gated CNFET with a 500 nm channel (Figure 1) is used first to compare the on-and off-current of large band gap D CNT ≈ 1.0 nm with small band gap D CNT ≈ 1.4 nm revealing 50x reduction in leakage current at V DS = −0.65 V and simultaneous 2.5x degradation in drive current at V DS = −0.05V for side-bonded Pd contact, consistent with previous results. [7,12]The same structure is used to demonstrate a self-aligned MoO X /Pd contact doping scheme for contact resistance improvement, which achieves 26% high drive current in 500 nm channel FETs with larger band gap CNTs without impact to sub-V T slope or I MIN .Temperature-dependent measurement confirms the doping mechanism to improve charge injection through lower "effective" Schottky barrier height, and systematic experimentation defines an effective process window for contact resistance (R C ) enhancement.The robustness to 200°C annealing and stability of the doping after 6 months is validated.Single-CNT FETs with 50 nm channel length are used to quantify the R C, which is reduced by 57% for the median device from 49 kΩ/CNT to 21 kΩ/CNT.Thus, this work reveals a promising doping method to enhance driving current in low-leakage CNFETs.

Results and Discussion
To compare the electrical properties of CNTs with smaller and larger band gaps, commercial CNT powders arc-discharge (Carbon Solutions, Inc, AP-CNT, diameter 1.3-1.7 nm), HiPCO (ATOM, HiPCO, diameter 0.8-1.3nm) were enriched by polymer sorting methods to form high purity semiconducting CNT solutions, as described in supplementary information section one.The CNT diameter distribution was further confirmed by absorption spectra, cross-section TEM, and AFM, Figure S1, with the median CNT diameter of 1.41 nm and 1.0 nm obtained for arc-discharge CNT (AD-CNT) and HiPCO CNT (HP-CNT), corresponding to median E G of 0.6 eV and 0.85 eV respectively.The summary table of CNT diameter and bandgap is shown in Figure S1g.It is important to note that the electrical bandgap of carbon nanotubes can be affected by the external dielectric media. [13]For the purposes of this paper, we have simply reported the bandgap as E G = 0.85 eV/D CNT (nm) from tight-binding approximation, however, we do not resolve the exact bandgap by experiment since our main objective is to compare two populations of CNT powders with differing diameters.
In Figure 1 the electrical performances of CNTs were validated by back-gated random-network CNFETs with Pd contacts, which is currently the best-known p-type contact for CNFET. [4,14]The tradeoffs between small E G and large E G CNTs are shown in Figure 1 at low V DS to emphasize on-current trends in the absence of saturation effects and at high V DS to emphasize leakage current trends in a regime where the leakage is above the measurement noise floor.At high V DS = −0.65 V, 2 orders of magnitude reduction in I MIN was observed in the 1.0 nm diameter CNFET, validating that CNTs with a larger band gap are better suited for energy efficiency (Figure 1b).In these p-type back-gate Schottky-barrier FET devices with contact overlapping gate electrode, the onset of ambipolar electron current at metal-CNT junction sets I MIN , and larger E G suppresses tunneling through the Schottky barrier to conduction band for electrons (Figure 1c).It is important to note for MOSFET-like devices with doped exten-sions, a different leakage mechanism dominates from band-toband tunneling at extension-channel junction, but the leakage suppression from a larger band gap is similar. [7]In Figure 1d, a 2.5x reduction in the driving current at low V DS = -50 mV in 1.0 nm diameter CNFET was observed, confirming the performance penalty of larger-band gap CNTs.At low V DS, contact resistance is a larger portion of the overall CNFET resistance due to avoiding saturation effects, and the Figure 1e band diagram illustrates the increasing Schottky barrier for hole conduction at the contact.
To enhance the performance of HiPCO CNFET (HP-CNFET), a strong and stable doping method for CNT is desired, which can be effective beneath or adjacent to the contact metal.As a p-doping material, MoO 3 has demonstrated degenerate pdoping for low-dimensional materials such as 2D transition metal dichalcogenides (TMDs) and CNT. [15,16]Illustrative band diagrams are depicted in Figure 2a to explain the doping mechanism of MoO 3 , which is a material with a large electron affinity ranging from 5.5 -6.0 eV depending on its stoichiometry. [17]ompared to the target channel material CNT, which has a band gap centered around the Dirac point of graphene ≈4.5 eV [18] with diameter-dependent band gap ranging from 0.6 to 1.06 eV (D CNT = 1.4 ∼ 0.8 nm), so the conduction band edge of MoO 3 is lower than the valance band edge of the CNTs, resulting in the formation of a heterojunction with a broken-gap band alignment.Due to the Fermi level difference, electrons are induced in MoO 3, causing the conduction band of MoO 3 to bend below the Fermi level, and holes accumulate at the valence band edge of CNT for p-type doping.Degenerately p-doped CNFETs have been demonstrated with a full channel-capped MoO 3 layer in Ref. [15] and Figure S2, Supporting Information, which validates the MoO 3 as a powerful p-doping material for CNT.The drawback of losing gate control by channel doping can be resolved with local doping only in the contact region, which to the best of our knowledge, has not been previously demonstrated in CNFETs. [15] contact doping structure is demonstrated utilizing MoO X nanoparticles deposited by electron beam evaporation on top of CNT within the contact window just prior to the Pd contact evaporation, as shown in Figure 2b-d.The morphology and statistical data of MoO X nanoparticles on the randomly distributed CNT network measured by AFM are shown in Figure 2c and Figure S3, Supporting Information.For deposited thickness (as measured by crystal monitor) below 0.8 nm MoO X nanoparticles were observed, while a continuous film would be formed for thicker MoO X deposition.The cross-section TEM and EDX depth profile shown in Figure 2d confirms Mo presence only at the Pd/CNT interface.In Figure 2e, the stoichiometry of Mo: O is 1:2.7 by XPS, which is typical for evaporated alloy films not to preserve the exact stoichiometry of the source material.
The process flow for device fabrication begins with randomlyorientated CNT network deposited through the immersion method.The P + Si substrate was first coated with 10 nm of ALD HfO 2 , treated by low-power oxygen plasma for 5 min, and then immersed in a semiconducting CNT solution for 24 h.After the CNT network deposition with a density ≈ 20 CNTs/μm, the substrate was rinsed with toluene for 20 min to remove excess polymer and unwanted CNT bundles, then baked at 180 °C in air for 30 min to improve CNT adhesion.Next, the channel region was defined by e-beam lithography followed by O 2 plasma etching to remove CNTs outside the channel.Next, a bilayer resists PMMA A4 495k, and A2 950k are spun-coat, and the contacts are patterned by electron beam lithography.In the final step, contact is deposited by electron-beam evaporation of MoO X nanoparticles followed by Pd metal, with liftoff in RemoverPG.Notably, the MoO X /Pd contact scheme is a self-aligned process without extra patterning steps or device area penalties.The samples were then baked at 200 °C for 30 min in N 2 ambient to verify robustness.

Electrical Data
The doping effectiveness of sub-monolayer MoO X nanoparticles on the CNT was electrically validated by blanket channel doping in Supplementary Information Section Two and Figure S2, Supporting Information.The CNFETs showed degenerate p-doping behavior, whether MoO X was a continuous layer 3.2 nm or a nanoparticle form.The enhancement in CNFET current at V DS = −0.05V for MoO X /Pd contact compared to Pd contact is investigated in 500 nm channel CNFETs as a function of MoO X thickness in Figure 3a.For 0.8-1.4nm of MoO X there is up to 40% drain current increase observed due to contact doping.Greater than 2 nm thick MoO X films reveal the resistance penalty from conduction through MoO X eliminates the doping benefit or severely degrades the current as seen at 4.8 nm.For thinner 0.4 nm of MoO X the doping is not as effective.The transfer (drain current (I D ) versus gate voltage (V GS )) in Figure 3b,c with various V DS and output (drain current (I D ) versus drain voltage (V DS )) curves in Figure S4, Supporting Information, with and without contact doping show the consistent current enhancement of the optimal 0.8 nm MoO X on HP-CNTs.To fairly compare the device performances with identical channel carrier density, the transfer curves were plotted with the same overdrive voltage (V OV ).A modest increase in drive current is observed with a decrease in ambipolar current in the long-channel CNFET with MoO X contact doping.The benefit of contact doping on drive current can be larger in short-channel transistors in which contact resistance is a larger portion of the total device resistance, as we will explore later.The subthreshold slope remained unaffected by contact doping as our proposed method did not create any additional interface charges in the channel, as shown in Figure S5, Supporting Information.Importantly, the stability of MoO X contact doping was demonstrated in Figure S6, Supporting Information, which depicts the electrical properties of a MoO X contact-doped HP-CNFET immediately after the fabrication, after 200°C annealing in N 2 ambient, and after 180 days with no observed change in on-current.Without passivation MoO 3 readily interacts with moisture in the air, resulting in a decrease in the work function and degraded doping effect. [19,20,21,22]Several approaches in literature have been used to improve the stability of MoO X , including annealing in inert gas and the application of a passivation layer. [15,23]In this work, MoO X was effectively passivated by the Pd contact metal, which is both stable in air and inert to MoO X .

Doping Effect on Schottky Barrier
To deduce the mechanism for R C improvement, transfer curves are measured as a function of temperature to extract the Schottky barrier for holes in the HP-CNT network FET with and without MoO X contact doping.The simulated density of state (DOS) in the portion of CNT under Pd metal contact, the valence band profile near source contact, and the current spectrum in Figure 4c explain how MoO X doping affects the Schottky barrier for hole injection.Within a Schottky junction, charges can be transported from the metal to the semiconductor by either thermionic emission across a barrier or direct tunneling through the barrier, which depends on the dimensions of the Schottky barrier.The impact of p-doping the CNTs beneath the Pd contact metal (Figure 4a, region A) and the injection barrier for holes from the source to the channel (Figure 4a, region B) has been investigated by TCAD.The modeling assumes a carrier density of 2.8 × 10 −2 (nm −1 ) and 10 −1 (nm −1 ) for CNT with and without doping in the contact region.Figure 4b reveals that as the p-doping increases, the Fermi level of the CNTs in the contact region becomes closer to the valence band by ∼30 meV, consistent with the following Schottky barrier height extraction.This results in a decreased "effective" Schottky barrier height and enables holes to pass through from the source to the channel more efficiently.A 54% increase in the on-current can be calculated by integrating the current density in Figure 4c.To confirm this mechanism, the I D -V GS are measured in Figure 4d-e for undoped and doped CNFET between 220 K and 320 K, and V DS was fixed at -50 mV at a vacuum pressure of 1×10 −6 Torr.A simplified thermionic emission current density is given by: I D = A d A h T*exp(qΦ hole /kT), where A d is the drain contact area, A h is the effective Richardson's constant for holes, T is the temperature, k is the Boltzmann constant, and Φ hole is the hole SB height.The temperature exponent was set to 1 instead of 2 because the carrier's transport within a CNFET was a 1D system.The corresponding hole energy barrier versus V GS plot derived from the Arrhenius relation (ln (I D /T) versus 1/T) was shown in the inset of Figure 4d-e.The effective Schottky barrier height was determined by the transition from a thermionically dominated regime to a tunneling regime.The effective Schottky barrier height decreased from 48 meV without doping to 25 meV with doping for HP-CNFET.The Schottky barrier height was extracted for several devices with consistent results, as shown in Figure 4f, with an average value of ∼26 meV after doping.

Contact Resistance Extraction
The contact resistance (R C ) was quantified from a single-CNT local back-gate FET by the methodology proposed by Pitner et al. . [4]o fabricate a CNFET with a single HP-CNT as the channel, the CNT density was decreased to ∼1 CNT/μm by reducing the deposition time of the immersion process, and the channel width of 150 nm was patterned.The source and drain contacts were patterned by e-beam lithography with a 50 nm channel length followed by the same MoO X /Pd contact doping scheme in the previous section.The total resistance in a single-CNT FET can be defined as R total = 2R metal + 2R C + R channel , as shown in Figure 5a.The metal resistance R metal was individually measured by two probes accessing both ends of the metal line.The channel resistance (R channel ) was minimized by having a channel length of 50 nm and operating the CNFET at high overdrive voltages (V OV ) such that we could assume 2R C >> R channel .Therefore, the resulting measured R C ≤ (R total -2R metal )/2 is the upper bound of the true R C, assuming R channel is negligible.The transfer curves for the pristine and contact-doped devices are shown in Figure 5b, and the cumulative distribution function (CDF) of the extracted contact resistance is shown in Figure 5c.The median R C decreased by 57% from 49 to 21 kΩ/CNT in MoO X contact doped CNFET, which matches well with the simulation result of 54% improvement in on-current.It's worth noting that the best two data points of undoped devices can still reach sub-10 kOhm/CNT, which is close to doped devices.However, they exhibit 1-2 orders of magnitude higher n-branch current than the rest at V GS = 0 V.This suggests that the variation of R C is due to the wide diameter (bandgaps) distribution caused by the CNT sorting process (Figure S1, Supporting Information).
The R C for smaller bandgap arc-discharge CNT and larger bandgap CoMoCAT CNT has also been studied in supplementary information section seven.For AD-CNT with a smaller bandgap (D CNT ∼ 1.4 nm, E G ∼ 0.6 eV), no R C improvement is observed due to no Schottky barrier exists (Figure S7, Supporting Information).Interestingly, for CoMoCAT CNT with a larger bandgap (D CNT ∼ 0.8 nm, E G ∼ 1.06 eV), there is no improvement in R C distribution because the doping is weaker than HP-CNTs as shown in Figure S7b-c.

Conclusion
In conclusion, this work demonstrates a path to using larger band gap HP-CNTs as the channel material in low-leakage CNFETs, with self-aligned doped contact to overcome the performance penalty of larger Schottky barrier on contact resistance.The mechanism of operation and effective process window are verified in control experiments, and a 57% improvement in the contact resistance was quantified using the short-channel single-CNT transistors.Ultimately, for high-performance, low-leakage CMOS technology, a sub-10 kΩ per CNT contact resistance must be achieved for a 10 nm contact length and then reproduced for an n-type contact.Therefore, contact resistance remains a key fundamental challenge to address.Extending this doping scheme to a contact length of 10 nm presents potential challenges, the probability of containing at least one MoO X particle at the ultrascaled tube-contact region decreases (Supplementary information section eight).Future studies may expand on this method by testing doping strategies capable of stronger p-type doping, pursuing the complementary strategy for n-doping to show low resistance scaled n-type contacts, or potentially exploring doping methods that integrate better underneath contact metal such as heteroatom doping or atomic layer deposition.

Figure 1 .
Figure 1.On-state and off-state performance versus various band gaps of back-gate CNFETs.a) Device schematic of back-gate CNFET with large and small CNT diameter as channel material.b) I MIN comparison between AD-and HP-CNFETs at V DS = −0.65 V. c) Simulated off-state band diagram at the drain side showing wider tunneling barrier for larger band gap.d) Driving current comparison between AD-and HP-CNFETs at V DS = −0.05V. e) Simulated on-state band diagram at the source side showing larger Schottky barrier for larger band gap.

Figure 2 .
Figure 2. a) Illustrative band diagram for the MoO X doping mechanism.b) Schematic of various thicknesses of MoO X deposit under the Pd contact.A thick and continuous MoO X film contributes to extra resistance.c) AFM image of HiPCO CNT network covered with nanoparticle form of MoO X .d) TEM image and EDX line scan of the Pd/MoO X contact.e) XPS spectrum of the MoO X .The stoichiometry of Mo : O is 1:2.7.

Figure 3 .
Figure 3. a) The driving current enhancement ratio for the doped contact compared to undoped contact as a function of MoO X thickness at V DS of −0.05 V. Drive current of CNFET with doped contact (blue) and undoped contact (red) at b) V DS of −0.05 V and c) V DS of −0.65 V.

Figure 4 .
Figure 4. a) Illustration of CNT under Pd metal lead (region A) and CNT inside channel (region B). b) Simulated density of states for CNT under Pd metal lead.The valance band of doped CNT (blue) is closer to the Fermi level compared to undoped CNT (red).c) Band diagram and current density inside channel for CNFET with undoped (red) and doped (blue) contact.d) Temperature dependence measurement of HP-CNFET with undoped contact at V DS of −0.05 V. Inset: The extracted hole energy barrier = 48 meV.e) Temperature dependence measurement of a HP-CNFET with doped contact at V DS of −0.05 V. Inset: The extracted hole energy barrier = 25 meV.f) Schottky barrier distribution for CNFET with undoped (red) and doped (blue) contact.

Figure 5 .
Figure 5. R C extraction.a) The structure of a back-gate single CNT transistor and the R C extraction methodology.The total resistance can be defined as R total = 2R metal + 2R C + R CH .b) Transfer curves of single CNT transistor with undoped (red) contact and doped contact (blue).c) Cumulative distribution function (CDF) of R C extracted from single CNT transistor with undoped (12 devices) and doped (24 devices) contact.