Flash memory cells data loss caused by total ionizing dose and heavy ions

Transcription

1 Cent. Eur. J. Phys. 12(10) DOI: /s Central European Journal of Physics Flash memory cells data loss caused by total ionizing dose and heavy ions Research Article Andrey Petrov, Alexey Vasil ev, Anastasia Ulanova, Alexander Chumakov, Alexander Nikiforov National Research Nuclear University MEPhI, Kashirskoye shosse 31, Moscow, , Russian Federation Received 19 March 2014; accepted 07 June 2014 Abstract: The paper provides experimental results of flash memory loss data investigation. Possible mechanisms of charge loss from storage element are reviewed. We provide some guidelines for flash memory evaluation to space application. PACS (2008): y, Xx, Tt Keywords: flash memory single event upsets total dose Versita Sp. z o.o. 1. Introduction Due to its non-volatility and density, flash memory is a very attractive choice for data and code storage in harsh environment applications. However, ionizing radiation can damage the floating gate cells of memory and cause critical data loss. This paper provides experimental results showing the information loss caused by total dose and heavy ions and describes the tests which must be performed to characterize the capability of flash memory storage for harsh environment applications. Flash memory is based on transistors with a storage element (Figure 1). Poly-Si floating gates, silicon nitride (Figure 2), nanocrystals etc. can be used as a storage element. The information in flash memory is stored by putting charge on the storage element which leads to a agpet@spels.ru change in transistor threshold voltage. Characterisation curves for drain current as a function of gate voltage for transistors in programmed and erased states are shown in Figure 3. Recently it was believed that the flash memory cells were much less sensitive to ionizing radiation compared to control circuitry and charge pumps [2], but scaling of advanced flash memory has led to its sensitivity increasing. 2. Data loss caused by total ionizing dose Data loss in flash memory cells occurs due to loss of charge from the storage element. The possible mechanisms of floating gate charge loss are described by Snyder et al. [3] and the energy band diagram of a floating gate transistor in the programmed state is shown in Figure 4. There are three basic components of a charge loss mechanism: 725

2 Flash memory cells data loss caused by total ionizing dose and heavy ions CONTROL GATE SiO 2 SiO 2 (3) RADIATION OXIDE STORAGE ELEMENT OXIDE POLY-Si (2) n+ POLY-Si p-si SOURCE SUBSTRATE DRAIN (1) Figure 1. Schematics of a transistor with a storage element. Figure 4. Energy band diagram of floating gate transistor in programmed state and three basic components of charge loss mechanism. GATE 1,E+07 SELECTABLE SOURCE DRAIN STORAGE ELEMENT 0 1 Figure 2. Transistor with a nitride storage element. NONCONDUCTIVE NITRIDE SELECTABLE SOURCE DRAIN Number of errors 1,E+06 1,E+05 1,E+04 1,E+03 1,E+02 1,E+01 No errors 1 R/E/W 2 R/E/W 3 Read only 4 Read only 5 Read only Dose, krad 1. holes generated in oxides around the storage element and escaped recombination injected in floating gate and recombine with electrons, 2. holes traped in oxides, DRAIN CURRENT [A] ERASED V read V th PROGRAMMED CONTROL GATE VOLTAGE [V] Figure 3. Drain current vs gate voltage characterisation curves for transistors in programmed and erased states [1]. Figure 5. Number of errors vs TID (black dots â Ș only read test, white dots - read/erase/write test). 3. electron emissions over the floating gate/oxide barrier. We conducted two different total dose tests. In the first test we irradiated 4 Gb NAND flash memory at U-31/33 (SPELS) LINAC in X-ray mode. The flash memory was irradiated in read only mode as well as in read/erase/write mode. The first observed failure was the data loss in some memory cells in read only mode. Failure in read/erase/write mode took place at much higher total ionizing dose (TID) levels (Figure 5). These results are in good agreement with the results obtained in works [4, 5], which also provide theoretical aspects of information loss mechanisms in flash memory. To get a better understanding of the data loss mechanism we only investigated the specially designed test chips based on floating gate cells in the second test. In this investigation we irradiated five couples of floating gate transistors (one in the programmed state, the others in 726

3 Andrey Petrov et al. Programming window,v 4,5 4 3,5 3 2,5 2 Dose,rad (Si) After refresh 1, Before 3 4 refresh 0, E+00 5E+04 1E+05 2E+05 2E+05 3E+05 3E+05 calculated the SEU cross section and received its dependence vs linear energy transfer (LET) of heavy ions (Figure 7). There is no difference in the SEU cross section for bias and unbiased modes. This result can be explained by the fact that in a static mode transistors are in the same mode as in unbiased mode and electric fields in oxides are the same. There are two common models of charge loss in storage elements: 1. transient conductive path model [6], Figure 6. Programming window value vs dose level. erased states) and evaluated the programming window the difference in threshold voltage of the floating gate transistors in programmed and erased states. Only the programming window value was measured during irradiation. After irradiation the value of the programming window was measured and then sunjected to a refreshing procedure (transitor erased or programmed). The value of the programming window decreases with increasing TID level but is restored after the refreshing procedure (Figure 6). The programming window decrease can be attributed to the radiation induced floating gate charge loss causing a change in transistor threshold voltage. The result of this test shows that total dose causes a loss of charge stored in floating gates and changes in threshold voltages of floating gate transistors. This result explains the observed higher failure levels in read/erase/write mode because floating gate charge is restored in each write operation. Flash memory is widely used in harsh environment applications for critical boot data storage and our experimental results must be taken into account for correct flash memory behavior characterization. 3. Data loss caused by single energetic particles We performed protons and several heavy ion tests (Ne, Ar, Kr and Xe) of NOR flash memory manufactured by a 110 nm Spansion MirrorBitÂľ process. DUTs were programmed to all zeroes and increment patterns before irradiation. Irradiation was performed in biased static 3.3 V supply voltage mode and unbiased mode. For all the ions tests single event upsets (SEU) in the memory cells were observed. Importantly, all upsets were from 0 state (programmed) to 1 state (erased), thus only sensitive cells need to be considered in cross section calculation. We ve 2. transient Carrier Flux model [7]. The first model describes charge loss through a conductive path along the particle track. The floating discharge process in this model is viewed as an RC circuit active for times smaller than the initial recombination. R is the resistance of the leakage path, C the capacitance of the storage element. Charge loss from a floating gate described by equation (1). Q = Q o exp( t/rc), (1) where Q o the initial value of the charge for programmed cell. We know from published experimental data [8 10] and our own results, that R depends on heavy ion LET. This dependence can be determined by empirical equation [9]: ( ) n LETo R R o. (2) LET Parameters R o and LET o are unknown, but can be determined from our results and experimental data [11, 12]. Parameter n can vary from 1.5 to 2. To estimate these parameters, we need to know critical R defined as the value at which sufficient charge loss occurs for sense amplifier circuit of flash memory. Threshold levels of LET for SEU effects in cells of modern flash memory can vary from 1 to 10 MeV cm 2 /mg depending on technology. Estimates show that for 25 nm feature size flash memory critical R is 104 Ohm and for 100 nm critical R is 100 Ohm Ohm. This data can be approximated by equation (2) with estimated values of the parameters R0 and LET0: 20 Ohm and 20 MeV cm 2 /mg respectively. This model is good for estimating the SEU threshold in flash memory cells but this approach has difficulties with the physical interpretation of the transient conductive path mechanism. At another experiment, DUTs were irradiated by Ne ions in static mode with various supply voltages (Figure 8), control after irradiation was performed for three various supply voltages (2.7 V, 3.3 V and 3.6 V). 727

4 Flash memory cells data loss caused by total ionizing dose and heavy ions 1,E-08 1,E-09 1,E-10 1,E-11 1,E LET, MeV*sm 2 /mg 1,E-08 1,E-09 1,E-10 1,E-11 1,E LET, MeV*sm 2 /mg Figure 7. SEU cross section vs linear energy transfer (LET) of heavy ions in unbiased (left) and biased with supply voltage 3.3 V (right) modes. 1,3E-11 1,1E-11 9,0E-12 7,0E-12 5,0E-12 3,0E-12 1,0E Supply Voltage during irradiation, V 3,6 V 3,3 V 2,7 V Figure 8. SEU cross section for various supply voltages during irradiation and control. Similar to the previous test, there is no difference in the cross section for various supply voltages during irradiation. However, cross section increase was observed related to the increase of control supply voltage. This difference is due to the fact that in some cells stored charge is only partially lost after irradiation and its threshold voltage are very close to erased state voltage. So, in read mode the supply voltage is set on the control gate and with this voltage increase leads to a higher drain current which could be recognized as erased mode. Proton DUT irradiation was performed after TID exposure to the level up to 75 percent of failure one. DUTs were irradiated by protons in static biased mode (3.3 V supply voltage) and unbiased mode. We observed notable differences between TID irradiated and non-irradiated devices (Figure 9). This effect requires further investigation to be explained. Another model links charge loss to differences between in and out floating gate flux of carriers generated by particle strike. This model is developed for floating gate storage elements, but can be applied for nitride based storage elements [12]. 1.6e e e e e e e e Unbiased mode non Biased mode non Biased mode Figure 9. Distribution of SEU cross section for irradiated and nonirradiated devices Conclusion We have provided experiments and proposed guidelines for flash memory test mode selection during radiation experiments. Flash memory devices often store critical boot data and are not refreshed during space missions. Our results show that TID failure level in read only mode without refreshing can be much lower than the level with refreshing. If it is intended to use memory for data storage only it is necessary to evaluate devices in only read mode. It must be taken into account that the SEU cross section in unbiased and biased modes are similar but increases with supply voltage increase. Our experiments show that cross section increases with TID level too, which should be taken into account in space applications. Our charge loss estimation from the transient conductive path model shows an increase in cells SEU sensitivity for modern flash memory. But this model can t explain the physical processes in this case and requires further studies. 728

5 Andrey Petrov et al. References [1] S. Gerardin et al., IEEE T. Nucl. Sci. 60, 1953 (2013) [2] D. N. Nguyen, S. M. Guertin, G. M. Swift, A. H. Johnston, IEEE T. Nucl. Sci. 46, 1744 (1999) [3] E. S. Snyder, P. J. McWhirter, T. A. Dellin, J. D. Sweetman, IEEE T. Nucl. Sci. 36, 2131 (1989) [4] M. Vujisic, K. Stankovic, E. Dolicanin, P. Osmokrovic, Radiat. Eff. Defect. S., 165, 5, 362 (2010) [5] E. Dolicanin, Gamma ray effects on flash memory cell arrays, Nucl. Technol. Radiat. 27, 284 (2012) [6] G. Cellere, A. Paccagnella, A. Visconti, M. Bonanomi, A. Candelori, IEEE T. Nucl. Sci. 6, 3304 (2004) [7] N. Butt, M. Alam, Single Event Upsets in Floating Gate Memory Cells, Proc. Int. Reliab. Phys. Symp. (IRPS), Apr 2008, 547 (2008) [8] A. I. Chumakov et al., Radiat. Meas. 30, 547 (1999) [9] F. Irom, D. N. Nguyen, G. R. Allen, S. A.Zajac Scaling Effects in Highly Scaled Commercial Nonvolatile Flash Memories, IEEE Radiation Effects Data Workshop (REDW), 1 (IEEE, 2012) [10] T. R. Oldham et al., TID and SEE response of advanced Samsung and Micron 4G NAND flash memories for the NASA MMS mission, IEEE Radiation Effects Data Workshop (REDW), Quebec, Canada, 114 (IEEE, 2009) [11] G.Cellere, A.Paccagnella, L.Larcher, A.Chimenton, IEEE T. Nucl. Sci. 49, 547 (2002) [12] S. Gerardin, M. Bagatin, A. Paccagnella, A. Visconti, E. Greco, IEEE T. Nucl. Sci. 58, 827 (2011) 729