A review of yield modelling techniques for semiconductor manufacturing

Kumar, Naveen; Kennedy, Kathryn; Gildersleeve, K.; Abelson, Richard; Mastrangelo, Christina M.; Montgomery, Douglas C.

doi:10.1080/00207540600596874

Cited by 89 publications

(29 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The VDS model suggest that the bonding is caused by a point-like sources with a density of 0.9 mm −2 . These models and the other common yield models, 28,29,30 however, predict much higher probabilities than observed when the recess area is smaller than 1.5 mm 2 . This discrepancy is even higher with deeper recesses and the standard SC1 cleaning.…”

Section: Effect Of Temperature and Dilution Of Sc1 On Fusion Bonding-mentioning

confidence: 76%

Nondestructive Characterization of Fusion and Plasma Activated Wafer Bonding Using Mesa and Recess Structures

Varpula

Suni

Dekker

2014

ECS J. Solid State Sci. Technol.

View full text Add to dashboard Cite

We present two methods for characterization of wafer bonding. They are based on recess and mesa bond test structures with various shapes, measurement of unbonded regions using scanning acoustic microscopy (SAM), and image analysis. The first method maps locally the surface energy across the bonded wafers using the measured deformations around these structures and the finite element method (FEM). The FEM analysis is supported by analytical modeling. The second method uses the measured bonding probabilities of 10-19 nm deep recess bond test structures in investigation of surface interactions and in determination of the average of the surface energy at the wafer level. The present methods and proposed optimized test structures allow the evaluation of surface cleans without destructive, off-line methods such as the crack-opening method, which is employed as a reference. The methods are utilized in the investigation of the effect of O 2 and N 2 plasma activation and the dilution and temperature of Standard Clean 1 on Si/SiO 2 direct bonding. The results from both methods correlate with each other. The bond strength of the annealed wafers is observed to increase in the order 1) O 2 plasma, 2) standard SC1 at 65 • C, 3) N 2 plasma, and 4) dilute SC1 at 45 • Wafer bonding is an essential component of the semiconductor industry.1-3 It allows wafer level packaging, 3D integration of diverse devices, and manufacturing of silicon-on-insulator (SOI) wafers needed as substrates for microelectromechanical systems (MEMS) and high performance microelectronic and photonic devices.1 Perhaps the most common bonding process used, especially for SOI production, is fusion wafer bonding which usually follows the procedure:First the wafer surfaces are prepared in a cleaning treatment and then the wafers are joined together in vacuum at room temperature. The final bond strength is obtained during annealing at high temperatures. In plasma-activated wafer bonding 3,4 the wafers are pretreated in plasma. This allows high bond strengths to be obtained at lower annealing temperatures.Wet cleaning is an essential step required before fusion bonding of hydrophilic silicon wafers can succeed. The chemistry of Standard Clean 1 (SC1) is a convenient and highly effective method. Numerous early studies such as Refs. 5-8 all used SC1 as part of the standard wafer cleans before wafer bonding. There are three main objectives of the SC1 clean. Firstly, the importance of a hydrophilic surface terminated with hydroxyl groups, readily provided by a basic solution such as SC1, to bonding has long been noted by Maszara et al.,9 and elucidated in detail by Tong and Gösele.8 Secondly, the SC1 clean removes organic contaminants which are known to interfere with bonding reactions.6 Finally, the SC1 clean is very effective at removing particles from wafer surfaces as described in detail by Itano and Kezuka.10 As an additional criteria, roughening of the surface can be minimized with proper dilution of the cleaning solution.9,11 The SC1 clean meets these requirem...

show abstract

Section: Effect Of Temperature and Dilution Of Sc1 On Fusion Bonding-mentioning

confidence: 76%

Nondestructive Characterization of Fusion and Plasma Activated Wafer Bonding Using Mesa and Recess Structures

Varpula

Suni

Dekker

2014

ECS J. Solid State Sci. Technol.

View full text Add to dashboard Cite

show abstract

“…These models can be validated by testing their prediction errors and comparing them with other yield models. Several published yield models are described in [10,11] and are summarized in Table I. Stapper and Rosner [35] suggest that a value of = 2 for the negative binomial model gave the best results in their work over a 16-year period, so = 2 was used in computing the negative binomial predicted values in these comparisons.…”

Section: Die-level Model Validationmentioning

confidence: 99%

“…Many of these models are described in [11,25] and are briefly shown in Table I. In these models, the variables used are defined as follows: D 0 = defects per area D i = defects per area from process step i A = area of a die A w = area of a wafer n = number of processing steps = clustering parameter…”

Section: Modeling Approachesmentioning

confidence: 99%

See 1 more Smart Citation

Modeling and analyzing semiconductor yield with generalized linear mixed models

Krueger

Montgomery

2014

Appl Stoch Models Bus & Ind

Self Cite

View full text Add to dashboard Cite

As the challenges and opportunities of using 'big data' expand, there is a need to explore different ways of analyzing large datasets. The semiconductor industry is a good example of a manufacturing process where many data are collected throughout the fabrication of the product. These massive datasets are used for various purposes, primarily to detect problems and determine root causes, control the process, and build models that predict yield. The yield predictions are used for process planning, optimization, and control. However, many current approaches to yield modeling are limited because the actual processes violate the model assumptions, limiting the power of the models' use. This paper explores the use of generalized linear mixed models (GLMMs) to predict semiconductor yield and to provide significant information about the process using a large semiconductor yield dataset. Both batch-specific and population-averaged GLMM approaches are used and compared. Differences in link functions, sample sizes, and levels of aggregation (die-level and wafer-level models) are also compared with each other and with the results from generalized linear models (GLMs). The results of this study show that GLMMs are a reasonable approach to analyzing large datasets by providing additional insight into the fabrication process while maintaining or even improving prediction power compared with GLMs and some prior yield models found in the literature. This paper also provides a modeling strategy through suggestions regarding level of aggregation, link function, and sample size that are appropriate for different research goals.

show abstract

“…With the reduction in device sizes, re-entrant flows (repetition of similar processing steps), and the variety of products to be manufactured, the complexity has strongly increased these recent years. This complexity, combined with the strong competition, forces semiconductor manufacturers to introduce several layers of controls in order to guarantee high yield [1]. Therefore, after some process steps, control operations are introduced at different levels (product, process, and equipment) in order to verify that the process is still under control and the product within specifications.…”

Section: Introductionmentioning

confidence: 99%

A Literature Review on Sampling Techniques in Semiconductor Manufacturing

Nduhura-Munga

Rodriguez-Verjan

Dauzère‐Pérès

et al. 2013

IEEE Trans. Semicond. Manufact.

View full text Add to dashboard Cite

This paper reviews sampling techniques for inspection in semiconductor manufacturing. We discuss the strengths and weaknesses of techniques developed in the last last 20 years for excursion monitoring (when a process or machine falls out of specifications) and control. Sampling techniques are classified into three main groups: static, adaptive, and dynamic. For each group, a classification is performed per year, approach, and industrial deployment. A comparison between the groups indicates a complementarity strongly linked to the semiconductor environment. Benefits and drawbacks of each group are discussed, showing significant improvements from static to dynamic through adaptive sampling techniques. Dynamic sampling seems to be more appropriate for modern semiconductor plants.

show abstract

A review of yield modelling techniques for semiconductor manufacturing

Cited by 89 publications

References 42 publications

Nondestructive Characterization of Fusion and Plasma Activated Wafer Bonding Using Mesa and Recess Structures

Nondestructive Characterization of Fusion and Plasma Activated Wafer Bonding Using Mesa and Recess Structures

Modeling and analyzing semiconductor yield with generalized linear mixed models

A Literature Review on Sampling Techniques in Semiconductor Manufacturing

Contact Info

Product

Resources

About