Measurement Model Approach Research Articles

M3 - A hybrid measurement-modeling approach for CPU-bound applications on cross-platform architectures

Predicting performance of CPU intensive applications for a target platform is of significant importance to IT industries. However, the target hardware platform for which we want to predict the performance is often different from the testbed on which the application performance measurements are done, and may not be unavailable for deployment for various practical reasons. This paper presents M3, a Measure-Measure-Model method, which uses a pipeline of three steps to address this problem. The methodology starts with measuring CPU service demands of the application on the testbed. Then, it builds clones that mimic the application code in terms of the type of operations, the number and size of network calls with external servers and API calls made at different layers of the technology stack. The clones are simple and easy to deploy, yet demonstrate the same speedup factor between the source and the target as the original application. The clones are then deployed on the testbed and on the target to estimate the application CPU service demand under light load generation. In the final step, this estimated CPU service demand is fed into specific performance modeling tools that are capable of predicting application performance on the target under arbitrary higher workload. Predictions made using this approach are validated against direct measurements made on five applications in Java and PHP, and on a number of combinations of testbed and target platforms (Intel and AMD servers) and the prediction error is always less than 20%.

Micro-scale measurement and modeling of stress in silicon surrounding a tungsten-filled through-silicon via

The stress in silicon surrounding a tungsten-filled through-silicon via (TSV) is measured using confocal Raman microscopy line scans across the TSV both before and after etch removal of an oxide stack used as a mask to define the TSV during fabrication. Stress in the silicon arose in response to both athermal deposition and thermal expansion mismatch effects. The complex three-dimensional stress and strain field in silicon surrounding the TSV is modeled using finite element analysis, taking into account both athermal and thermal effects and the elastic anisotropy of silicon. Comparison of the measurements and model results shows that no one component of the stress tensor correlates with the Raman peak shift generated by the deformed silicon. An analysis is developed to predict the Raman shift in deformed silicon that takes into account all the components of the stress or strain tensor; the results of the model are then used as inputs to the analysis for direct comparison with measured peak shifts as a function of distance from the TSV. Good agreement between the measured and predicted peak shifts is obtained for the case of the intact oxide stack. A discrepancy between the measured and predicted shifts was observed adjacent to the TSV with the oxide stack removed; further modeling suggests the discrepancy is explained by the formation of a small void at the TSV-silicon interface during etching. The combined measurement-modeling approach serves to both validate the model, in this case for TSV design, and to extend the measurement capability of confocal Raman microscopy to complex stress fields.

Investigation on Variability in Metal-Gate Si Nanowire MOSFETs: Analysis of Variation Sources and Experimental Characterization

The characteristic variability in gate-all-around (GAA) Si nanowire (NW) metal-oxide-semiconductor field-effect transistors (SNWTs) is analyzed and experimentally investigated in this paper. First, the main variation sources in SNWTs are overviewed, with the detailed discussion on the specific sources of NW cross-sectional shape variation, random dopant fluctuation in NW source/drain extension regions and NW line-edge roughness (LER). Then, following the measurement-modeling approach, via calibrated statistical simulation that is based on the modified analytical model for GAA SNWTs with corrections of quantum effects and quasi-ballistic transport, the variability sources in SNWTs are experimentally extracted from the measured devices with 10-nm-diameter NW channels and TiN metal gate. The results indicate that NW radius variation and metal-gate work function variation dominate both the threshold voltage and on-current variations due to the ultrascaled dimensions and strong quantum effects of GAA NW structure. The NW LER also contributes, but relatively less, to the threshold voltage variation.

Error Analysis of the Impedance Response of PEM Fuel Cells

A measurement-model-based error structure analysis was applied to impedance data collected under galvanostatic modulation for a single polymer electrolyte membrane (PEM) fuel cell. The stochastic errors were found to be roughly independent of frequency and were similar for the two instruments used. The measurement-model approach was found to be capable of identifying both high-frequency and low-frequency artifacts in the spectra. The low-frequency inductive loops were found, in some cases, to be consistent with the Kramer–Kronig relations. In other cases, nonstationary phenomena were found to have influenced the low-frequency response. This work showed that once steady-state operation was achieved, the low-frequency inductive loops in the impedance response could be associated with physical phenomena within the fuel cell. In addition, the formalism of the measurement-model error analysis provides a means for determining whether a steady state has been achieved.

Integrated measurement-modeling approaches for evaluating residual stress using micromachined fixed-fixed beams

Two methodologies have been developed to determine the biaxial residual stress value in thin films using electrostatically actuated fixed-fixed beam test structures. In the first, we determine the compliance matrix of the support posts using 3-D finite-element analysis. The residual stress value is then found from the best fit between the measured and modeled deflection curves, with the residual stress as the only free parameter in the model. An accuracy of /spl plusmn/0.5 MPa for the average biaxial residual stress level is evaluated from the reproducibility of independent measurements over a wide range of loadings. The key to the second methodology lies in the recognition that for a given value of residual stress, there exists a unique family of deflection curves associated with two adjacent beams of different lengths. Therefore, compliance information can be extracted directly from the deflection curves. We proceed to show that essentially the same values of residual stress are found by the two methodologies, while the latter allows much more rapid extraction of the residual stress. With the second methodology established, we find that residual stress values vary across a quarter of a six-inch diameter wafer by 2.5 MPa for three structural levels of polycrystalline silicon in our five-level surface micromachining technology.

Apportionment of light scattering and hygroscopic growth to aerosol composition

During a recent campaign at the NOAA CMDL monitoring station on Sable Island, Canada (43.93° N, 60.01° W) a dual‐nephelometer humidigraph measured the hygroscopic growth factor of aerosol scattering, fRH(σsp), one of the key parameters necessary for estimating short‐wave aerosol radiative forcing. Measurements revealed less growth for anthropogenically influenced aerosols than for marine, fRH(σsp) of 1.7 ±0.1 vs. 2.7 ± 0.4, where fRH(σsp) = σsp(85%)/ σsp(40%). A combined measurement‐modeling approach was used to estimate σsp; and its RH‐dependence, based on the measured particle size distribution and composition. The model suggested that differences in the particle size distribution, assuming the same aerosol composition, could not explain the observed differences in fRH(σsp). We have confirmed with individual particle analysis, that aerosol composition was indeed responsible for the difference in fRH(σsp). As well, the scattering contribution of organic carbon for the influenced case is at least as much as sulfate aerosol.