Single Event Effects and Total Dose Test Results for TI TLK2711 Transceiver

Transcription

1 1 Single Event Effects and Total Dose Test Results for TI TLK2711 Transceiver R. Koga, Member, IEEE, P. Yu, and J. George Abstract-- TLK2711 transceivers belonging to the Class V dice manufactured by Texas Instruments were tested for their sensitivity to radiation. We measured single event effects as well as total ionizing dose effects. V I. INTRODUCTION ARIOUS types of COTS transceivers have been examined for their sensitivity to radiation [1]-[4]. The sensitivity may encompass bit errors as well as total ionizing dose effects. Recently, we carried out radiation testing of the Class V TLK2711 transceiver. Testing took place at the Lawrence Berkeley Laboratory (LBL) 88-inch cyclotron facility, the Radiation Effects Facility at TAMU, and the Indiana University Cyclotron Facility (IUCF). We utilized protons at LBL and IUCF, while heavy ions were made available at TAMU and LBL. Samples of heavy ions used for testing are shown in TABLE 1. Radiation tests are made up of (1) SEU (single event upset)/sel (single event latchup) testing with protons, (2) SEU/SEL testing with heavy ions, and (3) total ionizing dose (TID) testing with 50 MeV protons. TLK2711 is a 1.6 to 2.7 Gb/s transceiver (XCVR) manufactured by Texas Instruments [5]. It is also called a Serializer/Deserializer (SERDES). Our test samples were a part of the Class V dice. One group of samples was obtained in 2005, while another group of samples was obtained in A block diagram of TLK2711 is shown in Fig. 1. A device may be configured for a transmitter (TX) or a receiver (RX). It belongs to the TI CMOS (epi) 25C10 technology with a feature size of 0.25 µm [6]. Some of the features are listed below [5]: 2.5-V Power Supply; Low Power: <500 mw; 16-bit Parallel TTL (transistor to transistor logic) Compatible Data/Internally Encoded Into 20 Bits Using an 8- bit/10-bit (8b/10b) Encoding Format; Interfaces to Backplane, Copper Cables, or Optical Converters; On-chip PLL (phase lock loop) Provides Clock Synthesis From Low- Speed Reference; Receiver Differential Input Threshold 220 mv Minimum; Loss of Signal (LOS) Detection; and Integrated 50-Ohm Termination Resistors on RX. The 8b/10b encoding/decoding process utilizes a code, which maps a 8-bit symbol (8-bit word) to a 10-bit symbol Manuscript received July 10, R. Koga, P. Yu, and J. George are with The Aerospace Corporation, El Segundo, CA USA (telephone: , rocky.koga@aero.org All trademarks, service marks, and trade names are the property of their respective owners. (in TX) in order to achieve reasonable digital bit polarity (e.g., DC-balance without a high level of bits disparity) during the serial transmission and a reliable clock recovery (RXCLK) at the receiver (RX) as shown in Fig. 1 [7]. The clock recovery is accomplished with the use of PLL while the decoding of the 8b/10b code provides the 8-bit symbol at RX. Special symbols are utilized to designate such items as startof-frame, end-of-frame, etc. TLK2711 uses a comma symbol ( ) for byte alignment. TLK2711 latches 16-bit parallel data at the transmit data terminals (TXD0 to TXD15 at TX). In order to accomplish transmission, serialized bit stream (TXP/TXN and RXP/RXN) converted from two parallel 8-bit words at TX (8b/10b encoding) re-emerges as two parallel 8bit words at RX (8b/10b decoding) representing the original 16-bit parallel data. Possible sites for SEU as well as SET (single event transient) may take place in either the TX or RX section. Encoding and/or decoding would be affected when an SEU occurs at either place. It is conceivable that the PLL is a site where an SEU may originate or the site where its capability may be hampered by incoming low fidelity signals for proper clock signal reconstruction. SEE sensitivity of space-born microcircuits such as BER (bit-error-ratio) has been studied [BER = #errors/(bit transmitted) = σ.φ/(data rate)] regarding data links for satellite use [8]. In some cases BER has subsequently been referred to as bit-error-rate [4]. TABLE I CHARACTERISTICS OF HEAVY IONS Ions AMU Energy (MeV) LET [MeV/(mg/cm 2 ) Range in Si (µ) B O Ne Ne Ar Cu Kr Kr Xe Xe Note: Selected ions available at TAMU and LBL are listed. II. EXPERIMENTAL Test setups for all of our radiation tests were essentially identical. Each test setup is schematically presented in Fig. 2. Two test samples (one in the transmit mode while the other in the receive mode) were mounted on a PCB (printed circuit board). Both samples were exposed simultaneously during irradiation.

2 2 16 parallel bits TX D0 --TX D15 TXCLK TKLSB/TKMSB TX Parallel to serial conversion Two sets of 8b/10b encoding Encoded 10b serial, differential -TXP -- -TXN parallel bits RXD0 -- RXD15 RXCLK RKLSB/RKMSB LOS RX Serial to parallel conversion Two sets of 8b/10b decoding PLL & clock recovery Signal detection Encoded 10b serial, differential -RXP-- -RXN-- Additional Lines: RX receiving signals directly (without serializing) from TX (LOOPEN), pseudo random bit stream (PRBS) testing (PRBSEN), & other control lines for special k characters, etc. Fig. 1 Block diagram of TLK2711 (Serial/Differential 10b Encoded Data) BERT and Signal Generator (Serial 10b) TLK2711 XCVR (RX mode) (Parallel 8b Data and RX Clock Signal) TLK2711 XCVR (TX mode) Fig. 2. TLK2711 XCVR radiation testing setup. Test Board (Both RX and TX are Irradiated Simultaneously)

3 3 Fig. 3. Picture of test board. Fig. 4. Eye pattern of the TLK Gb/s output at pre-rad. level. (200 mv/div and 195 ps/div).

4 4 A test system made up of a signal generator and a BERT (bit error rate tester) was placed away from the PCB. A photo of the PCB is shown in Fig. 3. A thermistor to monitor the case temperature may be adhered with epoxy for each device on some PCBs. Signals (single-ended RX input mv; ref. to V ID, which is the receiver input voltage differential ) from the test system were led to the receiver on the PCB [5]. The parallel 8b data and the receiver clock signal were directed from the receiver to the transmitter on the board. The transmitter output (serial/differential 10b encoded data) was checked for upsets at the BERT. This made it possible for a real-time BER testing at a 2.5-Gb/s serial link rate. The serial link transmitted a continuous flow of packets each of which consisted of a 4-byte preamble and a 4096-byte frame followed by a 24-idle-byte string (24-byte long). Within one frame (4096 bytes) two bytes were used for the packet checksum. One byte was made up of one 8-bit word in the parallel domain. The checksum as well as any errors in each 10b word was used to identify upsets (non-burst errors). A burst error (a large number of upsets in the stream of 10b words) could be also detected. A burst error sometimes made it impossible at RX to re-construct the original bit pattern in the packet, and consequently a drop in the packet resulted. Therefore any event that was associated with a lost packet belonged to a sub-set of burst errors. Continuous power-on testing of test samples made it possible to monitor device bias current in real-time. This information could be used for latchup detection. The eye pattern is shown in Fig. 4 for preirradiation sample obtained in V OD (pp-p) [differential, peak to-peak output voltage with low pre-emphasis] is about 1,360 mvp-p. Jitter, which may be a result of a random variation in timing, etc, may be observed. The differential output signal rise (t r ) and fall (t f ) times are less than about 150 picoseconds. used several sets of devices for this group of tests. We did not expose the devices with protons with lower energies than 50 MeV. No SEL was detected. The upset cross-sections were obtained with the serial link of 2.5-Gb/s. Taking the number of transmitted bits in the system into consideration, we have calculated BERs for bit-error cross-sections as shown in Fig. 6. Fig. 5. Proton upset test results for TLK2711 XCVR. Test samples were obtained in III. RESULTS AND DISCUSSION A. Proton Induced SEU/SEL Sensitivity 50 MeV protons at LBL as well as higher energy ones at IUCF were used to measure SEU/SET (single event upsets/single event transients) and single event latchup (SEL). Some upsets were made up of a one-bit error (or a few-bit errors in some cases) in the stream of 10b words as detected by the BERT (please see Fig. 2.) We call these bit errors. The other type consisted of events with many upset bits (often larger than 100 bits, but normally less than about 1000 bits) in the stream of 10b words. We call these burst errors. Proton induced upsets are shown in Fig. 5. The tests were carried out with test samples obtained in Bit error cross-sections are shown with black diamonds, while burst error cross-sections are shown with white squares. The crosssections for burst errors were consistently larger than those for bit errors. It should be noted again that the number of bits in one burst is numerous. The cross-sections for upsets were measured while the two test samples (RX and TX) were irradiated simultaneously (thus, set in the vertical axis). We Fig. 6. Proton induced BER. Test samples were obtained in BER often depends on the bit rate with respect to the particle (e.g., proton) flux. An example of this phenomenon is shown in Fig. 7 for an ATTDA215B transmitter (silicon bipolar PECL device for fiber channel communication) operated at 1 Gb/s rate. BERs tend to increase (often drastically) when the

5 5 transmission rate reaches the critical point with respect to the particle flux. During data collections we have paid attention to the existence of micro pulsation of the beam flux. At LBL, for example, the pulse width is about a few nanoseconds every several tens of nanoseconds. At flux values close to what we utilized, the micro pulsation has minimum effect on our measurements. Test samples obtained earlier (~2005) showed similar responses to proton irradiations. The SEE sensitivity is shown in Fig. 8. The number of bit errors was comparable (within a factor of two) to that of burst errors. The upset cross-sections were obtained with the serial link of 2.5-Gb/s. Taking the number of transmitted bits in the system into consideration, we have calculated BERs for bit-error cross-sections. Relevant BER for these results are shown in Fig. 9. BER in Fig. 9 seems comparable to that in Fig. 6. Fig. 9. Proton induced BER. Test samples were obtained in B. Heavy Ion Induced SEU/SEL Sensitivity Fig. 7. Proton induced BER for ATTDA215B for various proton flux values. Heavy ions were used to measure SEU and SEL sensitivity of the transceiver to these ions available at TAMU and LBL. As in the protons cases, some upsets were made up of a onebit error (or a few-bit errors in some cases) in the stream of 10b words as detected by the BERT (please see Fig. 2.) We again call these bit errors. The other type consisted of events with many upset bits (often larger than 100 bits, but normally less than 1000 bits) in the stream of 10b words. We again call these burst errors. The number of burst errors consistently was larger than that of bit errors for each irradiation, as shown in Fig. 10. Fig. 10. Heavy ion upset test results for TLK2711 XCVR. Test samples were obtained in Fig. 8. Proton upset test results for TLK2711 XCVR. Test samples were obtained in The cross-sections were measured while the two test samples (RX and TX) were irradiated simultaneously (thus, set in the vertical axis). We used two devices (SN9 and SN12 obtained in 2005) for this group of tests. No SEL was

6 6 detected. For SEL tests with these samples the case temperature was about 60 degrees (Centigrade), while the fluence was about 1 x 10 6 particles/cm 2 at the LET value of 60 MeV/(mg/cm 2 ). In order to extend SEL testing range, we plan to test samples obtained in 2007 at higher temperatures with increased values of fluence. No LOS was detected during irradiation. C. Total Ionizing Dose Sensitivity 1) BER vs. TID TID sensitivity was measured for SN1 (obtained in 2005) while observing the functionality of the test samples as well as BER. BER is defined as the ratio of the number of upset bits to the number of transmitted bits as defined before [8]. We used only non-burst errors for this analysis. (An analysis that involves burst errors is shown below in section C-2.) The flux of the 50 MeV proton was about 5.5 rd (Si)/sec. We applied a 20 krd (Si) of dose, and measured degradation of the sample before applying additional 20 krd (Si). For each 20 krd (Si) of dose (during which about 3 x 10 8 packets were compared), we measured the number of upsets. No upsets were observed except for the 100 krd (Si) window as shown in Fig. 11. TID effects were observed when the TID was between 120 and 140 krd (Si). This was characterized by the functional degradation of the transceiver as well as eye pattern measurements. No major bias current change took place. For another sample obtained in 2007, no measurable TID effects were detected at 140 krd (Si). At 140 krd (Si), an eye pattern measurement was taken as shown in Fig. 12. V OD (pp-p) [differential, peak to-peak output voltage with low pre-emphasis] is 1,340 mvp-p. Jitter, which may be a result of a random variation in timing, etc, may be observed. The differential output signal rise (t r ) and fall (t f ) times are less than 150 picoseconds. 2) The Number of Dropped Packets vs. TID TID sensitivity was measured for SN1 obtained in 2005 while observing the number of dropped packets. A burst error sometimes made it impossible to re-construct the original bit pattern in the packet, and consequently a drop in the packet resulted. Therefore any event that was associated with a dropped packet belonged to a sub-set of burst errors. We used only burst errors for this analysis. During this run all burst errors ended up in a dropped packet. (An analysis that involves non-burst errors is shown above in section C-1.) The flux of the 50 MeV proton was about 5.5 rd (Si)/sec. For each 20 krd (Si) of dose (during which about 3 x 10 8 packets were compared), we measured the number of dropped packets as shown in Fig. 13. TID effects were observed when the TID was between 120 and 140 krd (Si). This was characterized by the functional degradation of the transceiver as well as eye pattern measurements. No major bias current change took place. IV. CONCLUSION The Class V TLK2711 devices were tested for single event effects as well as proton induced total dose effects. Single event upsets are made up of two types of errors: (1) one-bit error (or a few-bit errors in some cases) in the stream of 10b words and (2) a burst of errors with many upset bits in the stream of 10b words. A burst of errors sometimes made it impossible to re-construct the original bit pattern in the packet and consequently a drop in the packet resulted. The proton induced total dose limit was beyond 100 krd (Si). This is based on results including (1) functional and (2) eye pattern measurements. (We have not included parametric test results with gamma ray irradiation.) These results show that this device type is less sensitive to SEU effects for both heavy ions and protons than some device types of the previous generation (e.g., ATTDA215B). No single event latchup was detected up to the LET value near 60 MeV/(mg/cm 2 ). Devices such as VSC7216 (fiber optic transceiver) and TI SN65LVDS95/96 (LVDS SERDES transmitter) have shown to be SEL immune at this LET range (Sept, 2000 testing with VSC7216 and May 2007 testing with SN65LVDS95/96) [9,10]. We plan to test for SEL with TLK2711 beyond this LET range in the near future. No LOS was detected during irradiation. Fig. 11. BER vs. TID for TLK2711 XCVR. Test sample was obtained in 2005.

7 7 Fig. 12. Eye pattern of the TLK Gb/s output after 140 krd (Si) radiation. (200 mv/div and 195 ps/div). [4] Hamilton, B.J. and T.L. Turflinger, Total dose testing of 10-bit low voltage differential signal (LVDS) serializer and deserializer, IEEE REDW, pp , [5] TLK2711 (1.6 to 2.7 Gb/s TRANCEIVER), SLLS501--September 2001, Texas Instruments, Dallas, Texas. [6] TI radiation data, RadCharRdmp.pdfC. [7] Widmer, A.X. and P.A. Franaszek, A DC-balanced, partitionedblock, 8b/10b transmission code, The IBM Journal of Research and Development, 27, pp , [8] Marshall, P.W., et al., Space radiation effects in high performance fiber optic data links for satellite data management, IEEE Trans. Nucl. Sci., 43, pp , [9] VSC7216 Low Power 1.25Gb/s Transceiver, PB-VSC , Vitesse, Camarillo, CA, [10] SN65LVDS95-Q1 LVDS SERDES transmitter, SGLS207A-October 2003, Texas Instruments, Dallas. TX. Fig. 13. The number of dropped packets vs. TID for TLK2711 XCVR. Test sample was obtained in V. [1] [2] [3] REFERENCES Carts, M.A., et al., Single event test methodology and test results of commercial gigabit per second fiber channel hardware, IEEE Trans. Nucl., 44, pp , LaBel, K.A., et al., Single event effect results for candidate spacecraft electronics, IEEE REDW, pp14-21, Koga, R., et al., Comparison of heavy ion and proton-induced single event effects sensitivities, IEEE Trans. Nucl. Sci., 49, pp , 2002.