1. Introduction

In recent years, the importance of non-destructive analysis techniques for fault location in electronic packages has dramatically increased. Due to the steadily growing complexity of integrated circuit (IC) technology, e.g. 3D through silicon via (3D-TSV) technology [1], parallel enhancement of current inspection technology such as time-domain reflectometry (TDR) [2], X-ray [3] or scanning acoustic microscopy [4] in terms of sensitivity, resolution and measurement speed is urgently needed. Among the non-destructive approaches TDR is considered as an exceptionally fast method for fault detection. Current all-electronic TDR systems – also dubbed as “closed-loop radar” [5] – employ as the main components a step or pulse generator and a high-bandwidth oscilloscope. The electromagnetic signal is transmitted through a coax cable and a probe-tip to the device under test (DUT). Every impedance discontinuity in the DUT causes a part of the injected signal to be reflected. By monitoring these reflections in the time-domain it is possible to detect structural defects and distinguish very quickly functional from defective structures. In simple (linear) DUT structures it is also straightforward to determine the spatial location of the defect by signal velocity considerations. Initially, TDR has been developed for fault location in large electrical systems such as cables but was quickly adopted for failure analysis in smaller structures on printed circuit board level, in packages and interconnects.

The spatial resolution of a TDR system in terms of the minimum distance d_min required between two objects for their discrimination is directly linked to the rise time τ_rise of the injected signal by the relation [6] $d_{\min }={\tau }_{rise}\cdot c/4\sqrt{{\epsilon }_{r,eff}}$, with c being the speed of light and ε_r,eff is the relative effective permittivity of the transmission line. As visualized by the diagram in Fig. 1 , the maximum resolution of currently available high-end all-electronic systems with a measured τ_rise = 11.1 ps is limited to approx. 365 µm for ε_r,eff = 5 [7]. In order to go considerably beyond this limit and to allow the inspection of smaller scale chip-level structures alternative optoelectronic approaches are required.

 

Fig. 1 Minimum distance d_min between neighboring resolvable discontinuities vs. signal rise time τ_rise.

Download Full Size | PDF

The large potential of femtosecond (fs) laser-driven optoelectronic measurement techniques for the characterization of high-frequency devices was recognized very early – more than two-decades ago [8,9] – and has been investigated by many research groups worldwide [10–12]. For ultra-fast generation and sampling of electromagnetic signals electrooptic [8] or photoconductive (PC) [13] processes triggered by short optical pulses have been apply very successfully. Over the past years fs-lasers became increasingly robust, compact, user-friendly and cost-efficient which is now fostering an increased technology transfer from lab to real world applications.

TDR-based fault isolation using a fs-laser-driven optoelectronic system was presented only recently by the Intel Corp [14]. With this system a τ_rise = 5.7 ps has been achieved, leading to a substantial improvement over all-electronic systems. However, the main bottleneck within this system – still preventing to take full advantage of the principally available sub-ps optoelectronic switching speeds – is given by the components interconnected between the optoelectronic chip and the DUT: a waveguide and a probe-tip having a bandwidth of only 110 GHz. In this work, a technical solution for the effective elimination of this bottleneck is presented. Using advanced micro-machined photoconductive probes we demonstrate the optoelectronic TDR-based localization of a waveguide discontinuity with an unprecedented rise time of τ_rise = 1.1 ps measured for the reflected signal after 5.8 mm of DUT-internal propagation. In contrast to standard configurations, the presented TDR-measurements are made in a contact-free mode using capacitive probe/waveguide coupling. This approach appears not only attractive under device protection and automation considerations but also for direct fault localization through spatio-temporally resolved field mappings of suspicious chip regions. Our approach can hence be considered as a multidimensional extension of the standard TDR analysis concept.

2. Experimental set-up

The optoelectronic system, schematically sketched in Fig. 2(a) , is based on a classic pump/probe set-up for time-domain measurements. The main components are a Ti:Sapphire fs-laser with a center wavelength of 780 nm, 150 fs pulse width (FWHM) and 78 MHz repetition rate and two micro-machined identical PC probe-tips. Technically related models of such probe-tips with modified designs have been recently successfully applied for other research tasks as the THz near-field detection at optically excited graphite flakes [15] or silicon-based nanophotonic structures [16] as well as for the direct evanescent field sampling of THz modes propagating though porous fiber waveguides [17]. Here, we investigate for the first time their application in a TDR-scheme for pulse injection and sampling at thin-film microstrip (TFSM) lines used as DUT.

 

Fig. 2 (a) Schematic of the applied optical pump/probe set-up for terahertz time-domain measurements. (b) SEM picture of the tip region of the used probe-tips. (c) Detailed to scale model of the applied probe configuration and orientation used for the measurements.

Download Full Size | PDF

The TDR measurements are conducted as follows: Using a beam splitter (BS) the output beam of the fs-laser is split into a pump and a probe beam, both with an average power of 4 mW. The pump beam is focused on the active region of the first probe used for pulse injection (PI) into the DUT. The active region contains a pair of biased PC switches made of two 5 µm wide gaps between two Ti/Au-based branch lines and a continuous center line running to the tip apex. The planar electrode structures are integrated on a 1.3-µm-thin cantilever of PC GaAs-material grown by molecular-beam epitaxy at low-temperatures [18]. Fabrication details have been described earlier [19]. A SEM picture of the tip apex and a drawing to scale of the geometrical design and the configuration of the probe-tips on the DUT during the measurement are shown in Figs. 2(b) and 2(c), respectively.

Both probe-tips are positioned 10 µm above the TFMS line. The probe-to-probe tip distance d_ed = 1mm is kept fixed throughout all measurements. Pulses generated at the pair of PC switches (symmetrically biased against the center line) are guided along the center line with a Sommerfeld-mode-like radial field distribution [20]. The signal is capacitively coupled from the probe-tip to the TFMS line. From the coupling position the pulse is propagating in two anti-parallel directions along the TFMS line. The second probe-tip (marked PD for “pulse detection” in Fig. 2(a)) is placed above the TFMS line section being in shorter distance to an open-end. With this probe the injected and reflected signal is sampled in time-domain. A linear mechanical stage controlling the optical length of the probe beam is used to sweep the time-delay of the optical probe pulses. The sampling process can be understood as the inverted analog of the pulse excitation process: A part of the signal propagating along the TFMS line is coupled to the detection probe-tip. It is guided along the center line to the pair of PC switches. The detected photocurrent across the PC switches is linear proportional to the electric field sampled by the PC elements. To increase the signal-to-noise ratio a rectangular modulated bias voltage (3 V peak-to-peak_, f = 700 Hz) and lock-in amplifier detection is used.

3. Time-domain reflectometry measurements

In order to investigate the capability of the system for TDR-based fault isolation we consider as the DUT a TFMS line with known transmission characteristics in the THz-range [21]. This TFMS design has been previously applied for THz bio-sensing applications [22]. It is fabricated on a silicon substrate covered by a 500-nm-thick Ti/Au-based ground layer and a 57-µm-thick polypropylene/co-polymer acrylate film as dielectric layer. The signal line on top of the dielectric has a width of 31 µm and is made of the same metal layer system as used for the ground layer.

Figure 3(a) shows a time-domain signal recorded for a probe to open-end distance d_do = 2.9 mm. Using numerical field simulations the coupling loss α_c for a vertical probe-to-sample distance d_pp = 1-10µm is calculated to be in the range of 18-22 dB. Still, a surprisingly large signal-to-noise ratio of the time-domain signal of approx. 900:1 is obtained for a lock-in amplifier averaging time of 30 ms. The signal trace can be divided in three characteristic parts: For t < −4 ps a pre-pulse caused by free-space radiation from the PI probe is detected. In the range of t = −4ps – 18 ps a pulse with increased amplitude is observed which represents the guided incoming signal propagating from the PI probe towards the open-end (signal in). For t > 18 ps the detected signal is dominated by the reflection from the open-end (signal out). As expected for an open-end reflection the polarity of the incoming and the reflected signal is identical. The peak amplitude of the reflected signal is around 51% (−6 dB) of the incoming peak amplitude. Considering the previously determined waveguide attenuation of ca. 1.0 dB/mm at 0.5 THz [21], this result indicates a nearly ideal reflection behavior with negligible radiation loss.

 

Fig. 3 (a) Measured time-domain signal for a vertical probe-tip to sample distance of 10 µm. The detection tip is placed in a distance of 2.9 mm from the open-end of the waveguide. (b) Time-domain data of the incoming and (c) the reflected pulse peaks measured at different probe-to-discontinuity distances d_do as listed. (d) Measured time-position of the reflected signal against waveguide shift from the open-end discontinuity. Fitting against a linear evolution reveals a location accuracy of ± 0.55 µm.

Download Full Size | PDF

An ultra-short rise time of τ_rise = 1.1 ps (determined as usual at the 10%-90% points of the rising signal edge) is measured which is an order of magnitude shorter than achievable with state-of-the-art all-electronic systems and should allow the differentiation of discontinuities separated by less than 36 µm (for ε_r,eff = 5).

Beside the ability to resolve neighbouring defects, even more important in many situations is the ability to accurately locate the position of a single defect. Beside short rise-times this also requires time jitter effects to be as low as possible. The latter point is a particular strength of optical coherent pump-probe measurement schemes [10] over all-electronic sampling systems, where time jitter values on the order of a few ps are common. We have quantified the location accuracy of our system by taking TDR scans at varying probe-to-end distances in a range of d_do = 2875µm – 2900 µm using 5-µm-steps. As visualized by the data in Fig. 3(b) the time position and slope of the incoming signal are virtually constant. This is expected since the transmission path between pulse injection and detection in forward direction is constant in this comparison, as long as microscopic material or structure defects can be neglected. We determine a residual time base error of approx. ± 5 fs for the incoming signal. For the reflected signals (signal out) shown in Fig. 3(c), very uniform signal slopes are registered which are, however, shifted towards earlier times for decreasing d_do. Within the monitored relatively small range of propagation length variations (2Δd_do = 50µm), dispersion or attenuation effects are negligible and a linear dependency between the temporal position of the reflected signal and Δd_do can be assumed. In Fig. 3(d) this correlation is plotted considering the half maximum position of the rising signal edge vs. Δd_do. Here, the determined standard deviation from a linear devolution is 2.6 fs and the maximum deviation is ± 5.1 fs which translates into a spatial location accuracy of ± 0.55µm. Since the observed jitter of the incoming and the reflected signals are almost equally large, we conclude that the accuracy of the mechanical translation stage used for the positioning of the DUT is sufficient and that the observed residual jitter must be dominantly caused by laser intensity fluctuations. The average group velocity and effective permittivity which can be further extracted from the data set of Fig. 3(d) are v_g = 216.5 µm/ps and ε_r,eff = 1.92, respectively, which are also in excellent agreement with literature data [21].

3. Spatio-temporal imaging of discontinuity location

All-electronic TDR measurements are usually restricted to single or few-point measurements. The determination of the location of a fault which was recognized only in the time-domain signal through a deviation from a “golden unit” reference signal will be a highly demanding in complex DUT structures (with the occurrence of simultaneous reflections from distributed discontinuities). Using data analysis based on electromagnetic field simulations for the retrieval of arbitrary fault locations in complex structures appears to be still unrealistic because of the enormous computational effort. Additional (differential) spatio-temporally resolved mappings, however, could be a very efficient way to detect or at least narrow the range of possible fault locations. As mentioned above, the experiments in this study where all conducted in a contact-free mode rendering spatio-temporal measurements a straightforward extension of our system. In order to demonstrate the capability of such measurements we have mapped the reflected pulse signal (corresponding to the “signal out” pulse of Fig. 3(a)) at the immediate open-end region of the TFMS line at three time-delays as shown in Fig. 4 . The configuration of the probes and the whole set-up was kept unchanged with the only difference being the two-dimensional lateral movement of the DUT at fixed time-delays. The time of the left diagram showing the first fully evolved reflection at the open-end is arbitrarily set to 0 ps. The spatially resolved data show very well that the pulse injection and detection range given by the contact-free mode is strongly confined to the TFSM line region, whose lateral boundaries are marked by the dashed line. The location where the reflection is generated is clearly visible. Moving the DUT against both – PI and PD – probes as done in this preliminary configuration has two implications: First, the spatial expansion of the reflected pulse propagating in negative y-direction (as well as its propagation speed) appear to be compressed (reduced) by a factor of 2. The second implication is that we do see a background signal in terms of a negative amplitude contribution beside the line which is a result of the modified transmission conditions under week PI coupling to the TFMS line. Both can however be avoided by including a fixed position for the PI probe against the DUT as well as a fiber-coupled pump beam delivery, which will also allow for spatio-temporal mappings at arbitrary pump/probe dislocations.

 

Fig. 4 Spatial resolved signal mappings at different time-delays showing the propagation of the pulse reflection from the open-end of the TFMS line (x/y-axis alignment as shown in Fig. 2).

Download Full Size | PDF