The subject matter described herein relates generally to the field of semiconductor device manufacturing, more particularly to the failure of semiconductor devices due to electromigration failure, and most particularly to improved computational methods for determining susceptibility to electromigration failure of integrated circuit interconnect structures.
In essence, electromigration (EM) as used herein relates to the self-diffusion of metals along an interconnection (or “interconnect”) in a semiconductor device as high current density is passed through the interconnect. This metal migration can result in voids forming in some parts of the interconnect through the departure of metal atoms, and/or hillocks formed on other parts of the interconnect due to the accumulation of metal atoms. The presence of voids will increase the resistance of the interconnect, perhaps beyond the limits set by the designers for a properly functioning device, possibly resulting in device failure or degraded performance. Hillocks may be formed in a direction parallel to the plane of the semiconductor device, possibly resulting in punching through the separating dielectric and short-circuits if contact is made with an adjacent interconnect. Short-circuits can also result if a hillock develops in a vertical direction and punches through the inter-metal dielectric in that direction. Degraded device performance and/or failure is the expected result in either case.
Although several atomic-scale processes can lead to EM failures as noted above, the chief concern in practice generally relates to the development of open circuits due to void nucleation and growth, particularly under EM stress conditions with current densities, j, about 2 MA/cm2 (mega-amps/sq.cm.) at temperatures in the approximate range of 250-350 deg. C.
Electromigration has been recognized as an increasingly important cause of integrated circuit (IC) failure as thinner and more narrow metal films are used as interconnects, and current densities carried by the interconnects have increased. For a typical bulk metal, resistance heating and melting of the metal can be expected at a current density around 104 A/cm2 (amps per square centimeter of conductor cross-section), or about 2×106 A/cm2 for Cu. However, typical IC interconnects can sustain current densities of the order of 107 A/cm2 due in large part to the good thermal contact and heat transfer between the interconnect and the IC substrate, typically silicon. At such high current densities (or “stress” or “current stress”) EM is expected to play an increasing role in device failure.
In addition to the above, the thin film interconnects in ICs may possess small grain sizes in some cases and a high ratio of interconnect surface area (or interface area) to interconnect volume. This structure generally allows for many high mobility diffusion paths, facilitating the transport of metal atoms at reasonably low temperatures. If such diffusion paths having different mobilities intersect, accumulation of atoms (hillocks) or the nucleation of voids is expected to result at the intersection. Thus, this combination of a high driving force for electromigration (that is, high current density) and the availability of a network of different high mobility diffusion paths makes the interconnect particularly susceptible to EM damage.
EM damage is exacerbated by the modern trend towards smaller interconnect width and thickness, requiring the coalescence of fewer vacancies or voids in order to cause an unacceptable rise in resistance. In addition, the modern trend in IC fabrication employs smaller separation between interconnects separated by dielectric materials typically having relatively low dielectric constant (low k) and prone to punch-through. Thus, device failure due to hillock formation becomes a more serious concern.
As one author has stated:
Computer codes for modeling EM behavior are commercially available including “Sentaurus Interconnect, Version G-2012.06, June 2012” by Synopsys, Inc. (hereinafter “SI Code”). However, in view of the (potentially) millions of interconnects in a typical IC design, full simulations with a such a code would be prohibitively time consuming. In addition, such codes typically require numerous input parameters (perhaps as many as 30) to be supplied for the particular interconnect materials, geometries and surroundings under consideration, increasing the difficulty of a full interconnect network simulation with the SI Code, rendering it impractical to perform for many ICs of practical interest.
Thus, a need exists in the art for improved methods of identification of specific interconnects and locations within a candidate IC that are particularly susceptible to EM failures and, therefore, call for more detailed study either in the laboratory or by means of a detailed simulation such as SI Code. Such early diagnosis of prospective EM problems can greatly accelerate the design and practical fabrication of reliable integrated circuits.
Accordingly, methods are described for performing detailed Technology Computer Aided Design (TCAD)simulations of electromigration (EM) failure in a standard test structure suitable for the simulation of integrated circuit (IC) conductive interconnects. The test structure described contains both vias and interconnects, is capable of simulating currents in both forward and reverse directions, and has symmetry allowing for reduced processing time in performing full TCAD simulations.
Methods are also described for performing a set of TCAD simulations and extracting therefrom those criteria that underlie EM failure lifetimes. Such criteria are much simpler and faster to implement than are the full TCAD simulations. Specific criteria are presented for Cu and Cu alloy interconnects under conditions reasonably realistic for operating ICs.
Methods are described for performing these simulations so as to extract from the results of these simulations criteria substantially underlying the EM lifetime of interconnects, thereby permitting rapid diagnosis of potential sites of EM failure early in the IC design and fabrication process, and thereby allowing more rapid development of reliable ICs robust against EM failure. Specific results for EM failure criteria in Cu interconnects are also presented.
These and other features will be understood upon consideration of the following detailed description of and the accompanying drawings.
To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures. The drawings herein are schematic, not to scale and the relative dimensions of various elements in the drawings are not to scale.
The teachings herein can readily be understood by considering the following detailed description in conjunction with the accompanying drawings, in which:
After considering the following description, those skilled in the art will clearly realize that the present teachings can be readily used for early and rapid identification of possible electromigration (EM) failure modes and failure sites in conductive interconnects in integrated circuits (ICs), as well as techniques for determining EM lifetime criteria that can be more readily implemented for the analysis of potential EM failures than can full computer simulations.
Designing and fabricating IC interconnects that are robust against EM failure becomes a more serious challenge as line widths are substantially reduced with successive advances in IC technology. To be concrete in our descriptions, we will focus on the specific example of copper (Cu) interconnects and Cu-alloy interconnects, For economy of language we use Cu to indicate either Cu or Cu-alloy materials. This is by way of illustration and not limitation in that the methods described herein can be utilized for other interconnect materials, as would be apparent to those having ordinary skills in the art. Also, Cu interconnects have considerable practical importance in present and future ICs so these examples are likely to be among the most important cases for practical IC fabrication.
The reduced scale of Cu interconnects results in smaller grain sizes in the Cu material, increased dependency on the barrier layer(s) as well as substantial atomic migration at the Cu-capping interface. These effects (among others) complicate carrying out TCAD (Technology Computer Aided Design) simulations by, for example, requiring use of process-dependent vacancy parameters. One goal of the methods described herein is to provide methods to elucidate (and to elucidate for some examples related to Cu interconnects), the relationship(s) between the internal microstructure of Cu-alloy interconnects and physical EM failure kinetics. Obtaining this information concerning Cu-line EM lifetime criteria in rapid, early simulations would enable the improvement of the density, performance and reliability of the overall IC chip. Also, such methods may allow for the development of effective EM test structures and Cu-alloy doping elements that may further improve the EM lifetimes.
The methods reported herein to determine Cu-line EM lifetimes make use of detailed TCAD simulations performed for an assumed standard 3-dimensional (3D) EM test chip. Unless otherwise noted, the simulations reported herein were carried out with the SI Code noted above. The standard 3D EM test structure used herein is chosen to be compatible with 2×nm node BEOL technology and is depicted in
This structure of
In addition to the ability to test EM stresses in both forward and reverse directions, the test structure of
The simulations are based on the EM full void growth model and include vacancy super-saturation in the Gibbs free energy (ΔG) calculation based on Rosenberg's void growth theory. (“Void Formation and Growth During Electromigration in Thin Films,” by R. Rosenberg and M. Ohring, J. Ap. Phys. Vol. 42, No. 13, pp. 5671, December 1971, the entire contents of which is incorporated herein by reference for all purposes.) A Cu-vacancy agglomeration induced under EM stress typically lowers ΔG to nucleate voids as depicted in
We use herein the following values for Cu-line EM parameters which are believed to typical:
A sensitivity analysis for tf is carried out with the SI Code for the listed parameters as depicted in Block A. In Blocks B and C, IVC and VFD values are set based upon the initially used oxygen impurity content (or concentration) and the segregation concentration of Cu-alloy atoms respectively for the given Cu-alloy under consideration, typically Cu-aluminum (CuAl) or Cu-manganese (CuMn) alloy.
Block D sets the Gibbs free energy parameter values for SED and GED. Some embodiments of the methods discussed herein use these two parameters as the chief “tuning parameters,” correlated with the degree of Cv agglomeration near M1-M2 via (for the reverse current case,
The IVC includes a sum of all initial vacancies that can be affected by the Cu-alloy scheme with dopants using PVD (physical vapor deposition) seeds, including manganese (Mn), aluminum (Al), cobalt (Co), magnesium (Mg), tin (Sn), zirconium (Zr), gold (Au), silver (Ag), palladium (Pd), iron (Fe), beryllium (Be), among others, as well as follow-up annealing conditions (NH3, N2/H2, among others), in the temperature range from approximately 150 deg. C to approximately 350 deg. C. Components of the IVC may be located at Cu grain interiors, Cu grain boundaries and the interfacial interface underneath capping layer materials (CoWP, CoSnP, CoWB, SiCN, SiN, among others). Another portion of IVC can be related to the oxygen contamination under the capping layer. A source of oxygen (O) and TaN oxidation can be the Co-carbonyl precursor used in the CVD (chemical vapor deposition) process. Therefore, it seems prudent to include possible O-related vacancies in IVC until an oxygen-free process is available for process calibration.
For the particular case of CuAl alloy, Al segregated near the capping layer can react with oxygen to form Al2O3 that can increase initial vacancy concentration, IVC.
For the case of CuMn alloy, Mn atom segregation enhances EM lifetime continuously with Mn alloy concentrations from 0% to about 4%, apparently independent of oxygen at the SiCN/Cu interface. This suggests that Mn segregation is driven by something other then oxygen as noted in “Electromigration Extendibility of Cu(Mn) Alloy-Seed Interconnects and Understanding the Fundamentals,” Proc. of IEEE IEDM, December 2012, the entire contents of which is incorporated herein by reference for all purposes. This implies an increase in the segregated metallic Mn preferentially at the capping layer interface.
IVC may also be affected by smaller grain size underneath the capping layer (for example, SiCN), at least (⅓) to (¼) the grain size of the Cu bulk. These effects can be worse for narrower line widths due to possible strain stress during barrier line formation.
Since IVC values may underlie the effects of alloy doping and oxygen contamination, IVC is thus likely to be one of the important process parameters, accordingly calling for calibration with full TCAD simulations.
Since IVC is closely related to back-end of line (BEOL) process conditions, IVC might be calibrated for each particular process of interest.
Vacancy transport takes place predominantly at the Cu-void surface and at grain boundaries as injected electrons have a wind-like effect, leading to Cu atom migration during a high current density EM stress. Thus, transport kinetics can be determined by the two major SED and GED parameters, although the current density (j) is the dominant stress factor for Cv as an initial EM condition.
Cu-line EM lifetime was simulated for a range of interconnect lengths and current densities (branch E of
When a new or different BEOL interconnect technology is introduced or under consideration (e.g., 22 nm node), one can use physically based TCAD simulation to define these 3 different EM regions for the technology node; Safety, Aware and Failure. Based on the jL product, EM safety and failure boundaries can be specified. In this particular case, jL less than about 2000 A/cm corresponds to the safety region and jL greater than about 9000 A/cm to the failure region, as shown in
On the F branch in
Therefore, this proposed methodology can provide a fast and practical prediction of EM lifetime criteria by the following sequential steps: sensitivity analysis of key EM parameter correlation, determination of saturated lifetime length to define EM-aware region, and extractions of temperature dependent Ea and (jL)c. By using the method, we are able to fully calibrate manufacturing data based on new BEOL Cu-alloy fabrication as well as production-level test information from OEM and/or foundry.
It would be advantageous to have a relatively simple and rapid method for computing reasonably correct EM lifetimes and the approximate “Safe,” “Aware” and “Failure” regions. A simplified 1-dimensional (1D) analysis of EM reliability physics is presented here as a first step in motivating such an approach. This avoids the full complexity of a numerical approach to 3D EM physics based on Cv continuity equations and the relationship of Cv and Cvoid that are used in the TCAD tool simulations. As such, it allows much more rapid simulations, and thus many more simulations, facilitating a more thorough evaluation of the parameter space encountered in practical fabrication of ICs with reduced susceptibility to EM failures.
(1) Referring to
(2) Considering the hypothesis of a rectangular void shape from the top capping layer to the bottom liner layer, Rstep can be proportional to Lv from Rstep=ρs*Lv/Ws, where Ws is the possible conduction cross section and ρs is the combined resistivity of Cu and barrier and Rstep is the step resistance between the barrier layer and the Cu line. These approximations can be applied to derive the present model for ρs/Ws given below.
(3) As the current passes through the wire, the length of the void space gradually increases until it reaches a critical value, Lv, at which the atoms of copper in this region are depleted. Rearranging the expression in paragraph (2) above yields Lv=Rstep*Ws/ρs. We thus obtain ρs/Ws=[(ρb/(tb*W+tb*2H))−(ρCu/H*W)].
(4) The Cu drift velocity νd is the product of mobility (μ) and electric field (E), where μ is eZDv/kT using the Einstein relationship as given in standard texts and also in the SI Code cited above. E can be expressed as ρj. Note that Z is the elementary charge of Cu, j is the current density, Dv is the vacancy diffusivity. These expressions yield: νd=μ*E=(eZDv/kT)*(ρj).
(5) The Cu drift velocity νd can be expressed in terms of the EM vacancy flux (Jv) of −νd*Cv assuming no temperature stress gradients and no hydrostatic stress gradients are present, where Cv is a uniform vacancy concentration. Thus, νd can be expressed as follows: νd=−Jv/Cv=(eZDv/kT)*(νCuj). Dv can conveniently taken to be the average diffusivity (Do exp(−Ea/kT)).
Finally, we have the desired expression for tf: tf=(kT*Rstep*Ws/ρs)/(eZDo*ρCu*j) which is proportional to Ws j−nexp(Ea/kT). As noted above in [0042], n is the current density exponent as also appearing in Black's Equation. We have n=−1 for a Cu-line and the EM life time is exponentially proportional to Ea.
Although various embodiments which incorporate the teachings of the subject matter described herein have been shown and described in detail herein, those skilled in the art can readily devise many other varied embodiments that still incorporate these teachings.
This application claims priority pursuant to 35 U.S.C. §119, §120, and/or §365 from provisional patent application Ser. No. 61/713,931 entitled “Determination of Cu-EM Lifetime Criteria Using Physically Based TCAD Simulations,” filed Oct. 15, 2012. The entire contents of the cited provisional patent application is incorporated herein by reference for all purposes.
Number | Date | Country | |
---|---|---|---|
61713931 | Oct 2012 | US |