Trade-Off Relationship of Size and Density of Platinum Nanocrystal in Nonvolatile Memory Characteristics

Background

Memories that store charge in discrete nodes, such as metal NCs, are candidates to replace polycrystalline silicon floating-gate memories because they localize charge and reduce interference between cells. Metals with high work functions, such as platinum (5.65 eV) and gold (5.1 eV), are expected to give stronger coupling, faster programming and longer retention than silicon NCs. Memory performance depends on both NC size and density, but these two parameters trade off against each other in a limited area and are difficult to control independently, so their combined effect has remained unclear.

Design and approach

Metal oxide semiconductor (MOS) capacitors were fabricated on p-type Si with a 4.5 nm-thick thermally grown tunneling SiO2 layer. Thin Pt wetting layers with initial thicknesses of 2, 4, 6 and 8 nm were sputtered on this oxide and converted into isolated NCs by rapid thermal annealing (RTA) in N2 at 800 °C for 30 s, so that the film thickness set both NC size and density. A 60 nm-thick SiO2 control oxide deposited by plasma-enhanced chemical vapor deposition (PE-CVD) at 300 °C and 200 nm-thick aluminum gates completed the devices. For analysis, the authors combined NC density and surface area with the natural logarithm of the electric field across the tunneling oxide (Eox), which governs the tunneling probability.

How it was evaluated

NC morphology was examined by field emission scanning electron microscopy (FE-SEM) and atomic force microscopy (AFM). Charging was measured by capacitance–voltage (C–V) curves at 1 MHz under ±20 V forward and backward sweeps and as a function of sweep voltage, and compared with a capacitor without NCs and with reported Si, Au and Pt NC memories. Retention at room temperature was measured by transient capacitance (C–t) after programming at +20 V for 3 s.

Key results

Increasing the initial Pt thickness from 2 to 8 nm increased the mean NC size from 15.1 to 55.2 nm (samples of 15.1, 26.7, 47.3 and 55.2 nm) and decreased the density. The abstract, introduction, Fig. 2 caption, Table I and conclusion give the density range as 5.6 × 10¹¹ to 3.2 × 10¹⁰ cm⁻², whereas one sentence of the Results text gives 5.2 × 10¹¹ to 3.2 × 10¹⁰ cm⁻². NCs from 2 and 4 nm films were hemispherical and well separated, while some NCs from 6 and 8 nm films merged by ripening and coalescence into elliptical shapes; films thicker than 10 nm did not form fully isolated NCs and were excluded.

All NC capacitors showed counterclockwise C–V hysteresis from electron injection into the NCs. At a 40 V sweep, the flatband voltage shift increased from 5.75 to 13.05 V with NC size, compared with about 0.4 V for a capacitor without NCs, and it increased exponentially with sweep voltage, consistent with Fowler–Nordheim (FN) tunneling; the increase was slower for smaller NCs. Against reported memories (Si NCs: 1.8 V at Eox = 3.57 MV/cm; Au NCs: 6.5 V at 1.88 MV/cm; Pt NCs: 3.2 V at 6.67 MV/cm), the 15.1 nm Pt NCs gave 5.75 V at 3.31 MV/cm.

Normalizing the flatband voltage shift by NC density and ln Eox gave a linear relationship with the surface area of a single hemispherical NC, which agreed with the theoretical estimate, and datasets from the reported Si, Au and Pt NC memories followed the same linear tendency. The calculated stored charge density ranged from 1.38 × 10¹² to 3.32 × 10¹² cm⁻², corresponding to approximately 3, 14, 76 and 118 electrons per NC for mean sizes of 15.1, 26.7, 47.3 and 55.2 nm.

Retention curves showed a fast decay, attributed to Coulomb repulsion between charges stored in a single NC, followed by a slow decay, attributed to lateral leakage through weak spots of the tunneling oxide. Larger NCs decayed more steeply, so smaller NCs retained charge better; the 15.1 nm NCs gave 520 s, compared with 100 s for Si NCs and 90 s for germanium (Ge) NCs in earlier reports. The authors concluded that large NCs stored more charge but held it for a shorter time.

Open questions

For the quantitative analysis, the authors assumed that the Pt NCs were uniformly distributed and hemispherical, although NCs formed from 6 and 8 nm films included merged, elliptical particles with larger size variation.

How it connects to other research

The silicon nanocrystal memory of Tiwari and colleagues stores electrons in nanocrystals by direct tunneling, and its abstract notes that the limited size and capacitance of the nanocrystals limit the number of stored electrons; the present study varied the size of Pt NCs and related the stored charge per NC to its surface area and the tunneling-oxide field.