Resistance Switching Properties of Stoichiometric and Nitrogen Implanted Silicon Nitride Nanolayers on N and P-Type Si Substrates

This paper examines the resistive switching characteristics of LPCVD SiNx MNOS ReRAM cells on both heavily doped n- and p-type silicon substrates, focusing on the effects of nitrogen doping. Detailed comparisons of electrical properties through nitrogen implantation reveal variations in trap density and SET-RESET voltages between n and p conductivity Si substrates. Impedance spectroscopy further elucidates the conductive path formation and its resistance.

💡 Research Summary

This paper investigates the resistive switching behavior of silicon nitride (SiNₓ) based MNOS ReRAM cells fabricated by low‑pressure chemical vapor deposition (LPCVD) on heavily doped n‑type and p‑type silicon substrates. A 5.5 nm Si₃N₄ layer, deposited over a 2 nm thermally grown SiO₂ tunnel barrier, serves as the active dielectric. Two sets of devices were prepared: (1) stoichiometric SiNₓ (N/Si ≈ 1.33) and (2) nitrogen‑rich SiNₓ obtained by implanting 25 keV nitrogen ions at a dose of 1 × 10¹³ cm⁻². The implantation was performed on both n⁺ and p⁺ wafers, resulting in four distinct sample groups (SN5, SNI5, SP5, SPI5).

Structural analysis by high‑resolution cross‑sectional TEM shows that the SiNₓ layers remain amorphous after implantation, with atomically sharp interfaces to the Si substrate and no observable swelling or defect propagation into the bulk silicon.

Electrical characterization employed DC current‑voltage (I‑V) sweeps and impedance spectroscopy (IS). All devices exhibit forming‑free bipolar switching. For stoichiometric n‑type devices (SN5) the SET voltage (HRS→LRS) is ≈ +3.5 V, while p‑type (SP5) requires ≈ +4 V. Nitrogen‑rich devices need higher SET voltages: ≈ +4–5 V for n‑type (SNI5) and ≈ +5 V for p‑type (SPI5). RESET voltages are lower for p‑type devices (≈ –1.4 V) than for n‑type (≈ –3.4 V). The memory window (log (I_ON/I_OFF) at 0.1 V) exceeds three orders of magnitude for all samples. Current compliance (I_CC) was optimized at 100 µA for SET and 5 mA for RESET; values below these thresholds lead to unstable filament formation or incomplete RESET.

Trap density (Nₜ) was extracted from space‑charge‑limited conduction (SCLC) analysis using the transition voltage (V_TR) and trap‑filled limit voltage (V_TFL). Stoichiometric samples show higher Nₜ (1.4–2.5 × 10¹⁹ cm⁻³) than nitrogen‑rich ones (≈ 1.9 × 10¹⁹ cm⁻³ or lower). The second switching cycle generally exhibits increased trap density, attributed to residual defects generated during filament rupture.

Statistical analysis of SET/RESET voltage distributions across many devices reveals that n‑type cells have smaller relative standard deviations (σ/μ ≈ 0.09–0.11) compared to p‑type (σ/μ ≈ 0.15–0.41), indicating better voltage uniformity on n‑type substrates. The larger variability on p‑type is linked to higher contact resistance and interface non‑uniformity.

Impedance spectroscopy performed at 0.2 V bias in the low‑resistance state (LRS) yields Nyquist plots well fitted by a Randles circuit (Rₛ + (Rₚ‖Cₚ)). Rₛ (series resistance) ranges from ~80 Ω (n‑type stoichiometric) to ~285 Ω (p‑type). Rₚ (parallel resistance, representing filament resistance) is on the order of hundreds of kΩ for stoichiometric devices but drops to 40–127 kΩ for nitrogen‑rich devices, indicating that the filaments in N‑rich SiNₓ are more conductive. Cₚ (parallel capacitance, associated with the unswitched dielectric volume) lies between 50 pF and 66 pF, slightly higher for N‑rich samples due to a larger unswitched volume.

The combined electrical and structural data lead to several key conclusions:

1. Nitrogen implantation passivates Si dangling bonds, reducing trap density, which raises the electric field required for filament initiation (higher SET voltage) but simultaneously yields lower filament resistance (lower Rₚ).
2. n⁺ silicon substrates provide more uniform switching voltages and lower variability than p⁺ substrates, likely because of lower contact resistance and a more stable SiO₂/SiNₓ interface.
3. The choice of stoichiometry and nitrogen content allows tuning of the trade‑off between switching voltage, memory window, and filament resistance, offering a pathway to optimize SiNₓ‑based ReRAM for low‑power, high‑density CMOS‑compatible memory applications.

Overall, the work demonstrates that careful control of nitrogen doping and substrate doping type can be leveraged to engineer the resistive switching characteristics of SiNₓ ReRAM, supporting its integration into future non‑volatile memory architectures.