Abstract In recent years, two new techniques were introduced for flip chip debug; the Laser Voltage Probing (LVP) technique and Time Resolved Light Emission Microscopy (TRLEM). Both techniques utilize the silicon’s relative transparency to wavelengths longer than the band gap. This inherent wavelength limitation, together with the shrinking dimensions of modern CMOS devices, limit the capabilities of these tools. It is known that the optical resolution limits of the LVP and TRLEM techniques are bounded by the diffraction limit which is ~1um for both tools using standard optics. This limitation was reduced with the addition of immersion lens optics. Nevertheless, even with this improvement, shrinking transistor geometry is leading to increased acquisition time, and the overlapping effect between adjacent nodes remains a critical issue. The resolution limit is an order of magnitude above the device feature densities in the < 90nm era. The scaling down of transistor geometry is leading to the inevitable consequence where more than 50% of the transistors in 90nm process have widths smaller than 0.4um. The acquisition time of such nodes becomes unreasonably long. In order to examine nodes in a dense logic cuicuit, cross talk and convolution effects between neighboring signals also need to be considered. In this paper we will demonstrate the impact that these effects may have on modern design. In order to maintain the debug capability, with the currently available analytical tools for future technologies, conceptual modification of the FA process is required. This process should start on the IC design board where the VLSI designer should be familiar with FA constraints, and thus apply features that will enable enhanced FA capabilities to the circuit in hand during the electrical design or during the physical design stages. The necessity for reliable failure analysis in real-time should dictate that the designer of advanced VLSI blocks incorporates failure analysis constraints among other design rules. The purpose of this research is to supply the scientific basis for the optimal incorporation of design rules for optical probing in the < 90nm gate era. Circuit designers are usually familiar with the nodes in the design which are critical for debug, and the type of measurement (logic or DC level) they require. The designer should enable the measurement of these signals by applying certain circuit and physical constraints. The implementation of these constraints may be done at the cell level, the block level or during the integration. We will discuss the solutions, which should be considered in order to mitigate tool limitations, and also to enable their use for next generation processes.

Abstract Emission microscopy is usually implemented for static operating conditions of the DUT. Under dynamic operation it is nearly impossible to identify a failure out of the noisy background. In this paper we describe a simple technique that could be used in cases where the temporal location of the failure was identified however the physical location is not known or partially known. The technique was originally introduced to investigate IDDq failures (1) in order to investigate timing related issues with automated tester equipment. Ishii et al (2) improved the technique and coupled an emission microscope to the tester for functional failure analysis of DRAMs and logic LSIs. Using consecutive step-by-step tester halting coupled to a sensitive emission microscope, one is able detect the failure while it occurs. We will describe a failure analysis case in which marginal design and process variations combined to create contention at certain logic states. Since the failure occurred arbitrarily, the use of the traditional LVP, that requires a stable failure, misled the analysts. Furthermore, even if we used advanced tools as PICA, which was actually designed to locate such failures, we believe that there would have been little chance of observing the failure since the failure appeared only below 1.3V where the PICA tool has diminished photon detection sensitivity. For this case the step-by-step halting technique helped to isolate the failure location after a short round of measurements. With the use of logic simulations, the root cause of the failure was clear once the failing gate was known.