Closing the testing gap between CMOS and photonics Adam Osseiran National Networked TeleTest Facility Edith Cowan University, Joondalup, WA, 6027 Australia Abstract High speed circuits and systems in their vast majority are still invariably based on CMOS technology for obvious cost and power efficiency reasons. Photonics science is gaining momentum in these high speed systems and is traditionally found in telecommunications but also in high performance computing and recently in standard computing systems with the main objective of alleviating the CPU to memory bottleneck. RF processing is producing another level of complexity in the new emerging variety of embedded mixed-signals circuits and systems where a signal changes its media types several times without leaving the package assembly. This new paradigm where RF, analogue, digital and optical signals are integrated in one component entails an ad-hoc testing and hierarchical fault modelling approaches that embrace the individual approaches of each of these signal types and amalgamate them into one optimised approach. 1. Introduction Testing accounts for a significant fraction of an integrated circuit, boards or system manufacturing cost. The cost of testing digital, analogue, mixed-signal and radio-frequency (RF) circuitry is becoming particularly high due to the increased level of integration and the complexity of available testing equipment that covers the testing of such heterogenous circuitry. The recent increased demand for high-volume integrated photonic-electronic systems that meet high-yield industry standards has created the need for yet another testing technique capable of detecting new types of faults. The first part of this paper describes some new photonic systems and devices and presents one approach of using conventional electronic testing techniques for advanced heterogenous systems. One of the most recent uses of photonics is in high performance computing (HPC) systems that are constantly evolving and becoming faster and more sophisticated. However the electrical interconnection has become the bottleneck in high-speed signal processing systems . Owing to their immunity to EMI, low noise, transmission security, low loss and low weight, optical interconnects have been widely employed to realize high information capacity through parallel photonic signal processing . New platforms integrating photonics and microelectronics are featuring very high-speed and a broad spectrum of new applications. Integrated photonic circuit designs have been proposed with challenging approaches -. However with the emergence of these highperformance MicroPhotonics systems, it has become crucial to test their reliability and lifetime. The list of possible faults for a MicroPhotonic system is in fact very extensive and much more need to be considered; however by first taking into account the most likely faults, the development of a practical finite fault set becomes possible. In this paper, we investigate the parametric faults that commonly occur in photonic systems and demonstrate the ability of traditional electronic testing techniques to detect faults in photonic devices and systems. As a case study, an optical network header recognition circuit is examined and the output test waveforms due to the most common defects are experimentally evaluated.
2. Faults in Photonics There are many sources of faults in photonic systems depending on the nature of the devices used. As in mixed-signal devices, fault sources can be classified into two categories, namely, (i) hard or catastrophic faults and (ii) soft or parametric faults . Hard faults are mainly due to component permanent failure, such as the break of a fibre connection, or a failure in a laser source or a photodetector. These defects cause hard faults that can easily be isolated. Soft faults are caused by variations in component characteristics, which are greater than the tolerance limits. Examples include faults caused by the manufacturing process, the deterioration of a polariser due to manufacturing defects, the increase in coupling loss due to misalignment, and the reduction in amplification due to gain fluctuations. The first step in developing a realistic fault model is to review the variation types of defects and their causes and effects. In this paper we adapt the concepts of test matrix of analog parametric faults proposed by Sunter et al.  and extend it to develop test matrices for photonic systems. Failures in a photonic system can be summarised in the following scenarios: Scenario 1. Process parameters are within the manufacturing specification limit: In this case, the system operates within the allowed tolerance and is assumed to be functioning fault-free. However, if these parameters fall outside the manufacturing limits, they could be due to an oversight in design, such as insufficient phase margin causing, for example, oscillation for some combination of process parameters and extreme temperature. Scenario 2. Process deviation is outside the process specification, but does not cause any performance to fail its specification; this type of defect might eventually cause a failure and thus poses a reliability risk. An example is microscopic cracks on the fibre surface getting bigger when stretched or twisted, eventually causing the fibre to break. Scenario 3. Process deviation is outside the specified process limits and causes a performance to fail a specification. This category includes classic parametric faults (soft faults), such as a drop in the optical pump intensity causing the system to fail a required optical power level. Scenario 4. Process deviation is outside the specified process limits and causes a circuit to fail to function, such as an improper isolator before the Erbium-Doped Fibre Amplifiers (EDFAs) causing backward power and a system failure. Scenario 5. Any fibre breaks or changes in the circuit topology causing performance to marginally fail its specification and possibly generating a Fresnel reflection (the reflection that occurs at the ends of two fibres by differences in the refractive index between glass and air). This category includes classic catastrophic and hard fault. The focus in the example given at the end of this paper is on parametric faults being significantly more important since they are harder to detect, and dominate opens and breaks. 3. Quantification of Process Parameter Variations Since the major components used for most photonic systems are the transmitter, optical amplifier, optical channel and optical receiver, we will consider as an example of target Photonic RF link shown in Figure 1 and investigate the failure of the parameters that are relatively independent of one another but have the most significant impact on the performance of the system. Measured Bit-Error Rate (BER) is used to benchmark the system performance . The parameters chosen as sources of faults are optical power, wavelength, and the Instantaneous linewidth, for laser source; spectral range, and responsivity, for the receiver; and finally gain and wavelength range for the amplifier.
For accurate fault detection, variations in the absolute values of all these parameters as well as the difference between related parameters must be considered. Each fault must be quantified by determining the minimum process parameter variation that causes the fault. 4. Example: Optical Header Recognition In this paper, we consider the example of the optical header recognition structure shown in Figure 1 . This example illustrates the computation of all the test failures. A 4-bit packet is generated from pattern generator and modulated with 1550nm optical signal generated by a laser source. The modulated optical signal splits into 4 output fibre ports whose lengths are chosen to delay the modulated optical signals by T0, T0+T, T0+2T, and T0+3T, where T0 is the bit time of the input pattern. The photoreceiver array that integrates 4 discrete photodetectors, 4 variable-gain amplifiers and an RF combiner, with the amplifier gain are configured to match the input bit pattern. For example, for a Header 1011 we set the gain of the second amplifier to a low value which corresponds to a “0” state, while the gains of the other amplifiers are set to high values, which correspond to “1” states. The output electrical signal from the RF combiner was monitored by a digital oscilloscope.
Figure 1 Optical header recognition structure and the bit streams received by the photodetectors.
When the bit sequence of the header matches the gain of the photoreceivers the sum of the output signals from the photoreceivers produces the autocorrelation of the header bit sequence with a central lobe at N·T where N is the number of bits in the header and T is the bit time. By sampling the output autocorrelation waveform at N·T and compare the amplitude of the central lobe to a threshold level, a pattern match is generated. The ideal circuit process parameters used in this example are shown in Table 1. Single or multiple faults may occur in either the active or passive components of the circuit but in this paper we evaluate the circuit for different test sets, and only one parameter is assumed to vary at a time. This approach enables the detection of single defects. The voltage of the main lobe is between 3.5mV and 5mV and the voltage levels for side lobes are less than 1.8 mV. Table 2 shows the test sets and the fault detected. Note that the noise effect is not considered in this study. Process Parameter Laser wavelength Laser power Responsivity Amplifier output Pattern input
Specification 1550nm 1 mW 0.95 A/W 2 mW 1101000
Table 1. Process parameters
Test No. 1 2 3 4 5
Test parameters Ideal Pin<-10dBm Dresp= 0 Aout<1.2mV INpat= 1011000
Defect detected None Vout<1mv Thr < 3.5mV Vout<1mv No threshold
Table 2. Test set and fault detected and corresponding yield coverage and yield loss.
Sufficient number of measured output parameters is required for the computation of deviation of the defective components in the circuit. Therefore, simulation and experimental measurements have been carried out to study the effects of the injected faults on the behaviour of the circuit through observing the output waveforms. Figure 2 shows the simulation results for different test set, and figure 3 shows the experimental measurements. It’s clearly obvious that the simulation results match with the measurements.
Figure 2 Simulation and measurement results of each test set. (a) The process parameters are within the limit, no fault detected. (b) Laser power is less than -10dBm, fault detected. (c) One photodetector responsivity is zero, fault detected. (d) When 2 photodetectors’ Responsivity are zero, fault detected (e) One Amplifier’s output voltage is less than 1.2m V, fault detected. (f) When the input pattern changes to 1011000, fault detected.
The first test corresponds to the cases when all the process parameters are within the specification limit. In all cases no faults are detected and the optimum threshold is within the range 3.5mV
The last test illustrates the fault detection when a fault is added to the input pattern INPat changing it to 1011000. Figure 3(f) shows the output waveform for this fault test. Since the system operates fault-free for a 1101000 pattern, the error injected in the input pattern causes a significant change in the output waveform, and no threshold can be set in this case as only few sidelobes are generated. 5. Conclusion We have addressed some of the general defects that commonly take place in photonic systems. Broken or open fibres typically occur more frequently than parametric fault and have great consequences, but we focused on the process variations that cause parametric faults, more difficult to detect, and are dominant in comparison to breaks or open in fibres. An example of a faulty photonic system is illustrated by injecting faults in an optical header recognition system for different process parameters. The output waveforms give a good idea on how to detect a fault in a photonic system. More research is needed in this topic to cover integrated photonics systems in the same way as for analogue and digital systems. References  Mohammed Edris, et al, “Optical interconnection system integration for Ultra-short-reach applications” Intel Technology Journal, Vol.87, No. 2, pp.115-128, 2004.  Jaemin Shin, et al., “Comparison of electrical and optical interconnect”, Proc 53rd Electronic components and technology conf, pp 1067-1072, New Orleans, LA 2003.  Li-Ching Shen, et al. “Characterization of Organic Multi-mode optical waveguides for the Electro-optical printed circuit board (EOPCB), Proceedings of 6th Electronics, Packaging Technology Conference, pp. 753 – 758, Dec. 2004.  Bahaa E.A Saleh, Malvin Carl Teich, “Fundamental of photonics” john Wiley & sons, 1991.  Geert Van Steenberg, et al., “MT-Compatible Laser-Ablated Interconnections for Optical Printed Circuit Boards” Journal of Lightwave Technology, Vol. 22, No. 9, September 2004.  Han Seo Cho, et al., “High-Coupling-Efficiency Optical Interconnection Using a 90o Bent Fiber Array Connector in Optical Printed Circuit Boards”, IEEE Photonics Technology Letters, Vol. 17, No. 3,pp. 690-692, March 2005.  Byung Sup Rho, et al., “PCB-Compatible Optical Interconnection Using 45o Ended Connection Rods and Via-Holed Waveguides”, Journal of Lightwave Technology, Vol. 22, No. 9, pp. 21282133, September 2004.  Stephen R. Forrest, et al., “Integrated Photonics Using Asymmetric Twin-waveguide Structures”, International Conference on Indium Phosphide and Related Materials, Conference Proceedings, pp.13 – 16, May 2000.  Karim Arabi, Bozena Kaminska, “Parametric and Catastrophic Fault Coverage of Analog Circuits in Oscillation-Test Methodology” Proceedings of the 15th IEEE VLSI Test Symposium, pp 166171, April 1997.  Steve Sunter, Navena Nagi, “Test Metrics for Analog Parametric Faults” Proceedings of the 17th IEEE VLSI Test Symposium, pp.226 – 234. April 1999.  Kun-Yii Tu, et al., “Modeling Accuracy of a Fiber-Optics Test Bed Using MGF Method”, IEEE Photonics Technology Letters, Vol. 16, No. 12, pp.2646-2648, December 2004.  Muhsen Aljada and Kamal Alameh, “Integrated 10 Gb/s AWG-based correlator for multiwavelength optical header recognition”, Optics Express Journal, Vol. 16, Issue 7, pp. 5150-5157, March 2008.