Radiation Hardness Evaluation of the Analog Devices AD9042 ADC for use in the CMS Electromagnetic Calorimeter P. Denes, B. Lev, R. Wixted Physics Department, Princeton University, Princeton NJ 08544, USA 1 We report on the results of an irradiation study carried out on devices from all wafers in a lot of AD9042 40 MHz 12-bitADCs. These devices were irradiated in a 72 MeV proton beam in order to simulate the eects in CMS at the LHC. No change in dynamic performance was observed for doses up to 2 10 4 Gy along with 4 10 13 n=cm 2. The uniform behavior of all devices in the lot also indicates that radiation hardness can be assured by sampling tests on devices from one wafer in the lot. 1 Introduction The CMS Electromagnetic Calorimeter[1] will consist of 82 728 lead tungstate (PbWO 4 ) crystals arranged in a barrel and two endcaps. The scintillation light from the crystals is captured by a photodetector, amplied, digitized and transported by high-speed ber-optic links o the detector. The digitizing electronics on the detector must survive a total absorbed dose of 10 3 (10 4 ) Gy in the barrel (endcaps) along with a uence of 210 13 n=cm 2 in the barrel over the lifetime of the detector, which is dened to be the amount of time required to obtain an integrated luminosityof510 5 pb,1. The endcap photodetectors must operate up to 10 14 n=cm 2, however the digitizing electronics will be placed in such a way as to limit the total uence to 5 10 13 n=cm 2. The signal capture electronics consists of a custom monolithic wide dynamic range preamplier[2] and gain-multiplexing circuit[3] followed by the Analog Devices AD9042 40 MHz 12-bit ADC. The preamplier converts the photocurrent intoavoltage with a dynamic range of 16 bits. This voltage is then ampli- 1 This work supported in part by the U.S. Department of Energy under Grant DE-FG02-91ER40671 Preprint submitted to Elsevier Preprint 10 May 1998
ed by four clamping ampliers with gains of 1; 4; 8; 32. The gain-multiplexing circuit simultaneously samples each of these four voltages at the LHC bunch crossing frequency of 40 MHz. Voltage comparators and digital logic in the circuit then determine which of the four voltage samples is the largest (corresponding to the highest gain) below a certain saturation threshold. This (quasi-static) voltage is multiplexed and digitized by the AD9042. The output data consist of a oating-point representation of the data in the form G 2 G 1 D 12 D 11 D 10 D 9 D 8 D 7 D 6 D 5 D 4 D 3 D 2 D 1 where G i is a 2-bit code representing the gain range used (1; 4; 8; 32) and D i is the 12-bit ADC mantissa. In this way, the 12-bit ADC converts the full 16-bit dynamic range. Radiation hardness is a key concern for electronics at the LHC. The custom circuits being developed for the CMS Electromagnetic Calorimeter are fabricated in radiation-hard processes. The (commercial) AD9042 is fabricated in Analog Devices' proprietary XFCB 1.0 (extra Fast Complementary Bipolar) process. Although the intrinsic properties of the process and the design techniques used in the ADC result in a radiation-hard part, the part is not formally guaranteed to be radiation hard. In order to ensure that the ADCs for CMS meet the radiation hardness requirements listed above, CMS and Analog Devices (ADI) have dened the process ow shown in Figure 1. In this ow, lots of 18 wafers are fabricated and probed to determine functional die. The rst wafer is diced, and roughly 20 parts are packaged and tested under irradiation. If these parts perform as required after irradiation, the entire lot is accepted, otherwise the entire lot is rejected. The cycle time for the complete process is about 21 weeks, and about 20 cycles would be needed to provide ADCs for CMS. The process ow ensures radiation hard parts provided that the radiation hardness of all wafers in a lot is uniform. In order to prove this, we have performed tests on ADCs from one wafer lot, in a way which simulates the nal ow. A lot of 17 wafers of AD9042 was fabricated in April 1997. Twenty ADCs from the rst wafer were isolated and packaged. These ADCs were then tested under irradiation. After packaging of the remaining 16 wafers, ADCs were selected at random from each wafer such that sets of 32 ADCs per wafer were obtained. Of the 512 ADCs delivered, 4 from each wafer were reserved for future tests, and the remaining 28 ADCs per wafer were then tested under irradiation as described below. 2
ADI Fab Lot 18 wafers Probe entire Lot Dice 1 Wafer Pkg ~ 20 die Dice entire Lot CMS Test (Irradiate) Accept? Week 1-10 11-14 15 20 Delivered Parts 21 2 Test Description Fig. 1. ADC Process ow for CMS Calorimeter electronics in the LHC radiation environment will be subjected primarily to ionizing radiation and neutron uence. In order to simulate these eects simultaneously, we have performed irradiation measurements using the OPTIS proton beam at PSI[5]. The beam consists of 72 MeV protons, emitted over 10 cm 2. The nominal operating ux was 1:2510 9 p=cm 2 =s. At these energies, the proton energy deposition is ve times greater than for minimum ionizing particles. Similarly, the damage induced by one 72 MeV proton is equivalent to that of two 1 MeV neutrons. A two hour exposure at the nominal ux thus simulates the eects of a total dose of 10 4 Gy along with 2 10 13 n=cm 2, corresponding to the full lifetime of almost all of the CMS Electromagnetic Calorimeter. In CMS, the ADCs will sample the voltage waveform output of the oating point preamplier. The most natural way to test for eects due to irradiation would thus be to use simulated waveforms and try to quantify changes arising after irradiation. Although this approach would best illustrate the eects in CMS, it makes it more dicult to isolate and quantify the eects. For this reason, the sine waves of various frequencies and amplitudes were measured rather than preamplied pulses. The reference signal frequency for our measurements is 5 MHz, as this represents the bandwidth of the preamplier. 3
ADCs were irradiated either under full operating conditions or unbiased. Radiation induced eects appear to be slightly enhanced for unbiased devices. Unbiased devices were measured before and after irradiation, and operating devices were measured continuously during irradiation. (An advantage of proton beam irradiation is that the beam spot is well dened, so that additional electronics around the device under test need not be radiation hard.) The test arrangementisshown schematically in Figure 2. Sine waves were generated with an HP3335A synthesizer. The ADC was clocked with a dierential AC-coupled PECL clock running at a constant 40 MHz, and the resulting ADC data were stored in a 4 kilo-sample FIFO memory. As the CMS application is unipolar, DC coupling was simulated with the buer amplier shown in Figure 2. This arrangement also allowed dierent osets to be used for comparison.,q 5HI 0+] &ORFNV $'& Fig. 2. ADC measurement setup. A sine wave is DC-coupled to the ADC, which hsa AC-coupled dierential PECL clocks. The ADCs were characterized with 1.2, 2.5, 5.0 and 9.6 MHz sine waves, with full-scale amplitudes ranging from about 40 mv to 900 mv (the full-scale range of the ADC is 1 V). These sine waves were digitized at 40 MHz, and the results stored in the 4 kilo-sample buer. Normally, the ADC performance would be obtained by a fast Fourier transform (FFT) of the stored data. However, as the 40 MHz clock (from a quartz crystal) was not a precise multiple of the sine wave frequency, spectral leakage can obscure certain eects. In analyzing the data, therefore, a pure sine wave was rst t to the data, and an FFT was performed on the data with the t subtracted. The t thus determines the amplitude and oset of the sine wave, and the FFT was used to calculate the Signal-to-Noise Ratio (SNR) and other measures of dynamic performance. The uniformity of the conversion gains, derived before irradiation, is illustrated in Figures 3 and 4. Here, the gain of the ADC was obtained by using the t results for the dierent amplitudes of 5 MHz sine waves. The gain is shown in Figure 3 for the dierent wafers. The plotted value for each point represents the average of the 28 ADCs/wafer prior to irradiation, and the error bar represents the RMS of the distribution of amplitudes for the 28 ADCs. The 4
distribution when all ADCs are taken together is shown in Figure 4, with an RMS dispersion of 0.7% of the mean value. *DLQ :DIHU Fig. 3. Conversion gain at 5 MHz as a function of wafer number. The open squares represent the average of the ADCs prior to irradiation, and the error bars the RMS. 35 30 25 20 15 10 5 0 0.975 0.980 0.985 0.990 0.995 1.000 1.005 1.010 1.015 1.020 1.025 Gain Fig. 4. Conversion gain at 5 MHz for all ADCs, i.e. a histogram of the data in Figure 3. The RMS is 0.7% of the mean. 3 Results Of the 512 ADCs (32 ADCs per wafer 16 wafers) 416 were irradiated up to a total dose of 10 13 p=cm 2 (which simulates the full life of the barrel CMS Electromagnetic Calorimeter), 32 up to a total dose of 210 13 p=cm 2, and 32 up to a total dose of 4 10 13 p=cm 2. The SNR before and after irradiation is summarized in Table 1. The use of sockets, and the somewhat noisy accelerator environment, resulted in a typical SNR of about -63 db rather than the -67 db specied for the device[4]. As seen from Table 1, the average change in SNR for ADCs irradiated at 10 13 p=cm 2 was 0:0 1:2 db, for ADCs irradiated at 2 10 13 p=cm 2 was 0:1 1:0 5
Dose f SNR [db] p=cm 2 MHz Before After 10 13 1.2,63:1 0:8,63:2 0:8 2 10 13 1.2,63:1 0:8,63:2 0:7 4 10 13 1.2,63:0 0:8,61:6 1:3 10 13 2.5,63:3 0:9,63:4 1:1 2 10 13 2.5,63:5 0:8,63:7 0:8 4 10 13 2.5,63:5 0:6,62:1 1:6 10 13 5.0,63:5 1:1,63:7 1:0 2 10 13 5.0,63:6 0:6,63:7 0:9 4 10 13 5.0,63:7 0:7,62:2 1:4 10 13 9.6,63:4 0:6,63:3 0:7 2 10 13 9.6,63:3 0:7,63:4 0:6 4 10 13 9.6,63:3 0:6,62:0 1:1 Table 1 SNR for the ADCs before and after irradiation. For ADCs irradiated at db, and for ADCs irradiated at 4 10 13 p=cm 2 was 1:4 1:6 db. Thus no degradation in SNR was observed at doses up to 2 10 13 p=cm 2. At double that dose, a decrease in performance was apparent with a loss of 1:4 db in SNR. Similar eects were observed for the spurious free dynamic range, which is given by the worst harmonic (as a fraction of full-scale) in Table 2. As before, the behavior is similar at doses of 10 13 p=cm 2 and 2 10 13 p=cm 2. Also, a decrease in performance is observed at doses of 4 10 13 p=cm 2. In addition, a decrease in post-irradiation performance is observed for input frequencies above 5 MHz. These results are displayed graphically in Figures 5 and 6 which show the change in 5 MHz SNR for all devices irradiated up to doses of 10 13 p=cm 2.In Figure 5, the change after irradiation is plotted as a function of wafer, and shows the very similar behavior from wafer to wafer. The change for all wafers is shown in Figure 6, and displays a Gaussian shape. 6
1 2 3 4 5 6 7 8 9 Dose f Worst Spur [db FS ] p/cm 2 MHz Before After 10 13 1.2,78:8 2:6,79:0 2:7 2 10 13 1.2,79:6 1:7,79:2 2:7 4 10 13 1.2,80:1 2:6,75:2 3:6 10 13 2.5,77:8 2:6,78:1 2:5 2 10 13 2.5,78:9 1:7,78:1 3:1 4 10 13 2.5,78:2 2:7,75:4 3:7 10 13 5.0,77:8 2:8,76:1 2:6 2 10 13 5.0,77:7 2:6,75:9 2:8 4 10 13 5.0,78:4 3:1,74:9 3:4 10 13 9.6,79:1 2:9,76:7 1:9 2 10 13 9.6,79:7 2:8,76:6 1:7 4 10 13 9.6,78:9 2:4,75:7 3:8 Table 2 Worst spur (largest harmonic) for the ADCs before and after irradiation. Change in 5 MHz SNR [db] 10 8 6 4 2 0-2 -4-6 -8-10 10 12 13 14 15 16 18 Wafer Fig. 5. Change in 5 MHz SNR for ADCs exposed to 10 12 p=cm 2 as a function of wafer number. The open squares represent the average of the ADCs irradiated, and the error bars the RMS. 4 Discussion and Conclusions For doses of 10 13 p=cm 2, which correspond to 10 4 Gy along with 2 10 13 n=cm 2 there was no statistically signicant change in dynamic performance for the 416 ADCs evaluated at this dose. Similarly, the 32 ADCs irradiated to 2 10 13 p=cm 2 did not appear to degrade after irradiation. For the 32 ADCs irradiated to 410 13 p=cm, however, a small decrease in dynamic performance was observed. For static performance, all ADCs showed a change in conversion 7
0 20 40 60 80 100 120-10 -8-6 -4-2 0 2 4 6 8 Change in 5 MHz SNR [db] Fig. 6. Change in 5 MHz SNR for all ADCs exposed to 10 12 p=cm 2, i.e. a histogram of the data in 5 gain with irradiation. In order to elucidate this eect, ADCs were studied in detail during irradiation. Figure 7 shows an example of the change in amplitude with irradiation. A 5 MHz sine wave, near full scale, was measured during irradiation. The (normalized) amplitude shown in Figure 7 is the ratio of the ADC reading at the indicated dose, divided by the initial reading. As can be seen, the full scale of the ADC increased about 1% during the irradiation, so that the measured amplitude decreased by about 1%. Simultaneously, the value of the ADC internal voltage reference (that is used in setting the gain) was measured, and was found to increase about 20 mv after irradiation. 9 5HI >9@ ( ( ( ( ( (,QWHJUDWHG'RVH>SFP @ 1RUPDOL]HG$PSOLWXGH Fig. 7. Amplitude and voltage reference change during irradiation. The crosses indicate the measured value of the ADC voltage reference on the left-hand scale, and the triangles indicate the normalized change in amplitude for a 5 MHz full scale sine wave on the right-hand scal. The change in reference voltage and measured amplitude are completely correlated. If one "corrects" the data by adjusting the ADC scale by the measured reference voltage, one recovers a straight line for the amplitude, with no measurable slope. It is felt that the change in reference voltage, and thus the ADC gain, arises for the following reason: The voltage reference circuit makes use of 8
a temperature-compensated current that is created by a bipolar current mirror whose transistors have dierent emitter areas. With irradiation, the DC current gain of the bipolar transistors diminishes, and the resulting increase in base current causes an error in the reference current. While this eect is dicult to correct in the ADC, it is relatively easy to correct in CMS by including a special radiation-hard reference in the front-end gain-multiplexing circuit. The value of this reference would be periodically measured as an oline correction to the amplitude. This reference must also be temperature compensated, but the precise voltage value of the reference is not important, only that it is stable. We have measured roughly 500 AD9042 ADCs from all wafers in a lot and nd the pre- and post-irradiation properties of all ADCs to be essentially identical. At doses of 210 13 p=cm 2, which correspond to 210 4 Gy along with 410 13 n=cm 2, no signicant change in dynamic performance was observed. A small decrease in conversion gain was observed, which is completely correctable with the addition of some extra circuitry. The pre- and post-irradiation behavior of the AD9042 thus meet the electrical requirements for the CMS Electromagnetic Calorimeter. Finally, the very uniform characteristics of the ADCs from dierent wafers serve to validate the production process ow envisaged, thus ensuring that all ADCs maintain their performance throughout the life of the experiment. 5 Acknowledgments The authors wish to express their gratitude to Tom Gratzek and Frank Murden of Analog Devices for their help in the rapid implementation of these tests, and their continued technical assistance. In addition, the authors would like to acknowledge the support from their PSI colleagues and the accelerator operators at OPTIS throughout these tests. References [1] CMS Collaboration, The Electromagnetic Calorimeter Project Technical Design Report (CERN/LHCC 97-33, 1997). [2] J.-P. Walder, High Dynamic Range and Low Power Very-Front end Preamplier for CMS Calorimetry, in: Second Workshop on Electronics for LHC Experiments (Balatonfured, 1996) 61. 9
[3] P. Denes, CMS Electromagnetic Calorimeter Readout, in: Second Workshop on Electronics for LHC Experiments (Balatonfured, 1996) 48. [4] Analog Devices, Inc., AD9042 12-Bit, 41 MSPS Monolithic A/D Converter (1996). [5] PSI User's Guide (Paul Scherrer Institut, Villigen, Switzerland, 1994). 10