Silicide/Si Hetero-Nanocrystal Nonvolatile Flash Memory
Jianlin Liu,Zhao, Dengtao,Zhu, Yan
DOI: https://doi.org/10.1109/isdrs.2005.1595969
2005-01-01
Abstract:As one type of discrete storage nodes, Si nanocrystal has been extensively investigated as the floating gate in nonvolatile MOSFET memory devices [1]. The discrete nature offers the immune of the charge storage from the local weak point in the tunnel oxide, which has dramatically driven the memory device toward the target of being smaller, faster and less power consuming. However, the defect-related traps play a dominant role in the long retention mechanism, which in other words means the non-reliable device performance. Although this issue can be solved by using metal nanocrystals, new issue is introduced: some metal would react with tunnel oxide in the high-temperature annealing process. In this work, we propose to use metallic silicide/Si hetero-nanocrystal floating gate and prove from both simulation and experiment that such hetero-nanocrystals can significantly improve retention without depressing the programming speed. The metal/oxide inter-diffusion is also minimized, leading to better endurance. We use CoSi2/Si hetero-nanocrystals as an example to demonstrate the concept. Figure 1 shows the band structure of the memory devices with CoSi2/Si heteronanocrystals. It is evident that the hetero-nanocrystal offers an additional barrier for the stored electron, which leads to longer retention time. Fig.2 and Fig.3 give the calculated retention and writing/erasing performance of the CoSi2/Si hetero-nanocrystal memory devices. Obviously the retention quality is orders of magnitude improved as compared to a defect-free Si nanocrystal memory. One notices that this improvement is not at the cost of the writing/erasing speed, as shown in Fig.3. The CoSi2/Si hetero-nanocrystals were fabricated with the self-aligned silicide technique [2]. First, Si nanocrystals were deposited in a LPCVD furnace on ~ 5 nm thermally grown SiO2. Then the wafer was covered with ~ 3 nm Co followed by rapid thermal annealing at 700 °C and 750 °C to form CoSi2/Si hetero-nanocrystals and CoSi2 nanocrystals, respectively. After the metal removal in a selective etchant, discrete CoSi2/Si hetero-nanocrystals or CoSi2 nanocrystals were left on the tunnel oxide. 15 nm oxide as control oxide was then deposited followed by aluminum electrodes formed by lift-off technique. The resultant devices are MOS structures with nanocrystals embedded. The capacitance-voltage measurements before and after the writing operation were performed to characterize the memory effect, as shown in Fig. 4 where writing was realized by bias the Al gate with 10 V for 1 s. It is found that the C-V shift due to the charge injection is almost the same for all three MOS devices with Si, CoSi2 nanocrystals and CoSi2/Si heteronanocrystals, respectively. The similar C-V shifts indicate the similar charge injection efficiency for these devices. This is further confirmed in Fig. 5 where the C-V shift, or memory window as a function of the writing time is plotted. One finds the similar charge injection speed for the three MOS memory devices. The retention performance was found to be significantly different for different devices, as given in Fig.6. In these first devices, the tunneling oxide quality is poor, leading to the fast decay within the first 100 seconds for all devices. Nevertheless, it is easily found that CoSi2 nanocrystal device possesses a much longer retention than that of the Si nanocrystal device. A 1.0 V flat-band voltage shift can be expected after 10 years for a CoSi2 nanocrystal memory device. It is also found that CoSi2/Si hetero-nanocrystal MOS memory exhibits the longest retention. A flat-band voltage shift of 1.5 V is still available after 10 years of retention at zero gate voltage. In summary, MOS memory with CoSi2/Si hetero-nanocrystal floating gate was proposed and investigated. Both theoretical simulation and practical device performance show that CoSi2/Si hetero-nanocrystal memory devices have much longer retention time while also with almost the same writing/erasing speed, as compared with its Si nanocrystal and metallic CoSi2 nanocrystal counterparts. References [1] S. Tiwari, et al. Appl. Phys. Lett., vol. 68, pp. 1377-1379, 1996. [2] J. P. Gambino, and E. G. Colgan, Mater. Chem. Phys. 52, 99 (1998)