Observations in Characterizing a Commercial MNOS EEPROM for

Observations in Characterizing a Commercial MNOS EEPROM for Space E. E. King, R. C. Lacoe, G. Eng, and M. S. Leung The Aerospace Corporation El Segundo, CA (310) 336 -7898, everett. e. king@aero. org 2004 MAPLD International Conference Ronald Reagan Building and International Trade Center Washington, D. C. September 8 -10, 2004 King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 1 P/MAPLD 2004

Hitachi HN 58 C 1001 Specification • 128 k x 8 -bit (1 Mbit) Electrically Erasable and Programmable Non-Volatile Memory • Single 5 V Power Supply (-10 V programming voltage generated on chip) • Access Time 150 ns • Automatic Byte or Page (128 bytes) Write • Write Time <10 ms • 104 Erase/Write cycles • 10 year Data Retention 2 P/MAPLD 2004

MNOS Memory Transistor Operation • • • Presence or absence of charge trapped in the silicon nitride layer results in a change in Vth between two states defined as a ‘ 1’ and ‘ 0’ Electrons are injected (write a ‘ 1’) by setting Vg = 5 V with -10 V applied to the well N+ Source A ‘ 1’ is erased (write a ‘ 0’) by applying 5 V to the well and setting Vg to -10 V, which injects holes The transistor state is interrogated by detecting channel current for a Vg = 0 V Data loss is generally due to thermal emission of the trapped charge in the silicon nitride (Ea = 1. 1 e. V) King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 3 Gate 28 nm Si 3 N 4 P well 1. 6 nm Si. O 2 N+ Drain N substrate P/MAPLD 2004

Typical Space Application • Memory written at beginning of mission • No power applied until the memory is needed to re-initialize the system • No memory rewrite planned during the mission • Memory speed is not critical • Maximum temperature is 70 C • Data must be retained for 15 years King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 4 P/MAPLD 2004

Our Test Objectives • Determine data retention lifetime – Stress parts at high temperature to accelerate data loss – Stress at two temperatures to determine the activation energy for the data loss process – Use data to predict data retention lifetime at 70 C – Characterize effect of TID on data retention • Develop qualification test to screen potential early failures out of the production part population – Determine worst-case test parameters – Develop worst-case test pattern King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 5 P/MAPLD 2004

Accelerated Aging Test Plan • Pre-condition parts – Write checkerboard and inverse checkerboard 220 times – Soak at 150 C for at least 48 hours • Age at 175 and 200 C - all leads grounded – Epoxy package material verified to be OK to >250 C – Half population written with checkerboard/half with inverse checkerboard – Parts tested at short intervals to catch early/intermittent failures (every 4 hours @ 200 C, 22 hours @ 175 C) • Determine impact of TID on aging – Irradiate parts to 78 krad in 26 krad steps (Co-60) – Repeat aging test post-radiation to assess affect of TID King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 6 P/MAPLD 2004

Aging-Induced Push-Out of Read Access Time No. of Data Byte Errors • Thermal stress produces a push-out of the read access time • The access time returns to its initial distribution after rewrite • Push-out extrapolated to 150 ns to estimate time-to -failure • Aging data at 175 and 200 C used to determine activation energy King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 40000 Control-Q 2 553 hours 30000 951 hours 1036 hours 1305 hours Rewritten 20000 10000 0 70 75 80 85 90 95 100 CE Access Time (ns) Part Thermally Stressed at 200 C 7 P/MAPLD 2004

Data Retention Lifetime • Assume that access time push-out is caused by charge leakage off the memory transistor – Push-out also might be influenced by the initial charge stored on the transistor and variations in sense amps • Error distribution vs. access time at 226 hours @ 200 C is similar to that at 1305 hours @ 175 C – Corresponds to an activation energy of 1. 2 e. V – Estimated time-to-failure (150 ns) at 70 C is a few hundred years • Worst-case activation energy for charge detrapping in the MNOS structure found in the literature is 0. 35 e. V – Estimated time for access time to reach 150 ns at 70 C based on this activation energy and our access-time pushout data is about 50 years King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 8 P/MAPLD 2004

Results of TID Tests • No significant degradation in performance found after 78 krad(Si) irradiation (unbiased) • Aging experiments after TID at 175 and 200 C showed the same access time push-out as had been measured in the initial aging experiments to within about 2% • Caution: The on-chip high voltage bias generator fails at 10 -to-20 krad when irradiated under normal operating voltages King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 9 P/MAPLD 2004

Signature of Intermittent Readout Failure OE • Intermittent failure observed during 0 aging due to oscillations on data I/O 1 outputs 1 – Not clear that increasing readout I/O 2 access time will solve problem CE • Oscillations worse as Vdd is Normal Waveforms decreased OE • Oscillations disappear as temperature increased to 38 C I/O 1 • Oscillation occurs long before there is a data retention problem due to I/O 2 aging • Oscillations spontaneously disappear CE in minutes to hours King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . Output oscillations 10 P/MAPLD 2004

No. of Bit Errors Parametric Testing -- Effect of Ambient Temperature 40 Vdd = 5. 0 V 30 tce = 150 ns 20 10 0 0 20 40 60 80 Temperature (C) • Number of errors decrease as the ambient temperature is increased • Recommend testing at below room temperature to increase probability of detecting a ‘weak’ part King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 11 P/MAPLD 2004

Test Pattern Sensitivity • A bit failure was estimated to be 26, 000 times more likely to be a ‘ 0’ than a ‘ 1’ • The worst-case pattern resulting in an error was reading a byte with all ‘ 0’s after reading a byte with all ‘ 1’s (70% of failed patterns) Data Pattern Written 1111 0000 Data Read Example 1111 00111111 Error • Percentage of failed patterns decreased as the numbers of ‘ 1’s in the prior byte decreased King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 12 P/MAPLD 2004

Test Pattern Developed to Screen out Weak Parts • Pattern heavily populated with ‘ 0’s (over 98%) to increase probability of finding an early failure • Short sequence of patterns inserted periodically in the address space to detect pattern sensitive failures • Unique data byte inserted when each address bit first went high to ensure that each address bit toggled • New pattern estimated to be 300 times more likely to detect an error than a checkerboard pattern King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 13 P/MAPLD 2004

Conclusions • Data Retention Lifetime found suitable for space – Activation energy determined to be about 1. 2 e. V, consistent with manufacturers value – Worst-case data retention time estimated to be about 50 years • Found no TID problem up to 78 k rad(Si) for unbiased irradiation • Identified a small probability for early failure due to data output oscillation – Consider operating part at Vdd>5 V to eliminate oscillations – Oscillations also eliminated when operated somewhat above room temperature – Unclear that extending the readout access time can ensure correct data readout King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 14 P/MAPLD 2004

Conclusions (cont. ) • Qualification test developed to increase probability of identifying ‘weak’ parts – Include tests conducted under worst-case parametric conditions for intermittent failure • Low Vdd • Low Temperature – Use special test patterns • High percentage of ‘ 0’s • Data patterns that include transitions from words with high number of ‘ 1’s to words containing high number of ‘ 0’s for every sense amplifier – Recommend eliminating any part exhibiting behavior that is inconsistent with the general population King Permission to copy or reprint is not required, but appropriate credit must be given to The Aerospace Corporation for the generation of this material. . 15 P/MAPLD 2004