Fat-Tree Routing for Transit

The Transit network provides high-speed, low-latency, fault-tolerant interconnect for high-performance, multiprocessor computers. The basic connection scheme for Transit uses bidelta style, multistage networks to support up to 256 processors. Scaling to larger machines by simply extending the bidelta network topology will result in a uniform degradation of network latency between all processors. By employing a fat-tree network structure in larger systems, the network provides locality and universality properties which can help minimize the impact of scaling on network latency. This report details the topology and construction issues associated with integrating Transit routing technology into fat-tree interconnect topologies.

Reconfigurable Architectures for General-Purpose Computing

General-purpose computing devices allow us to (1) customize computation after fabrication and (2) conserve area by reusing expensive active circuitry for different functions in time. We define RP-space, a restricted domain of the general-purpose architectural space focussed on reconfigurable computing architectures. Two dominant features differentiate reconfigurable from special-purpose architectures and account for most of the area overhead associated with RP devices: (1) instructions which tell the device how to behave, and (2) flexible interconnect which supports task dependent dataflow between operations. We can characterize RP-space by the allocation and structure of these resources and compare the efficiencies of architectural points across broad application characteristics. Conventional FPGAs fall at one extreme end of this space and their efficiency ranges over two orders of magnitude across the space of application characteristics. Understanding RP-space and its consequences allows us to pick the best architecture for a task and to search for more robust design points in the space. Our DPGA, a fine- grained computing device which adds small, on-chip instruction memories to FPGAs is one such design point. For typical logic applications and finite- state machines, a DPGA can implement tasks in one-third the area of a traditional FPGA. TSFPGA, a variant of the DPGA which focuses on heavily time-switched interconnect, achieves circuit densities close to the DPGA, while reducing typical physical mapping times from hours to seconds. Rigid, fabrication-time organization of instruction resources significantly narrows the range of efficiency for conventional architectures. To avoid this performance brittleness, we developed MATRIX, the first architecture to defer the binding of instruction resources until run-time, allowing the application to organize resources according to its needs. Our focus MATRIX design point is based on an array of 8-bit ALU and register-file building blocks interconnected via a byte-wide network. With today's silicon, a single chip MATRIX array can deliver over 10 Gop/s (8-bit ops). On sample image processing tasks, we show that MATRIX yields 10-20x the computational density of conventional processors. Understanding the cost structure of RP-space helps us identify these intermediate architectural points and may provide useful insight more broadly in guiding our continual search for robust and efficient general-purpose computing structures.

A Multi-scale Model for Copper Dishing in Chemical-Mechanical Polishing

The present success in the manufacture of multi-layer interconnects in ultra-large-scale integration is largely due to the acceptable planarization capabilities of the chemical-mechanical polishing (CMP) process. In the past decade, copper has emerged as the preferred interconnect material. The greatest challenge in Cu CMP at present is the control of wafer surface non-uniformity at various scales. As the size of a wafer has increased to 300 mm, the wafer-level non-uniformity has assumed critical importance. Moreover, the pattern geometry in each die has become quite complex due to a wide range of feature sizes and multi-level structures. Therefore, it is important to develop a non-uniformity model that integrates wafer-, die- and feature-level variations into a unified, multi-scale dielectric erosion and Cu dishing model. In this paper, a systematic way of characterizing and modeling dishing in the single-step Cu CMP process is presented. The possible causes of dishing at each scale are identified in terms of several geometric and process parameters. The feature-scale pressure calculation based on the step-height at each polishing stage is introduced. The dishing model is based on pad elastic deformation and the evolving pattern geometry, and is integrated with the wafer- and die-level variations. Experimental and analytical means of determining the model parameters are outlined and the model is validated by polishing experiments on patterned wafers. Finally, practical approaches for minimizing Cu dishing are suggested.

A Multi-scale Model for Copper Dishing in Chemical-Mechanical Polishing

The present success in the manufacture of multi-layer interconnects in ultra-large-scale integration is largely due to the acceptable planarization capabilities of the chemical-mechanical polishing (CMP) process. In the past decade, copper has emerged as the preferred interconnect material. The greatest challenge in Cu CMP at present is the control of wafer surface non-uniformity at various scales. As the size of a wafer has increased to 300 mm, the wafer-level non-uniformity has assumed critical importance. Moreover, the pattern geometry in each die has become quite complex due to a wide range of feature sizes and multi-level structures. Therefore, it is important to develop a non-uniformity model that integrates wafer-, die- and feature-level variations into a unified, multi-scale dielectric erosion and Cu dishing model. In this paper, a systematic way of characterizing and modeling dishing in the single-step Cu CMP process is presented. The possible causes of dishing at each scale are identified in terms of several geometric and process parameters. The feature-scale pressure calculation based on the step-height at each polishing stage is introduced. The dishing model is based on pad elastic deformation and the evolving pattern geometry, and is integrated with the wafer- and die-level variations. Experimental and analytical means of determining the model parameters are outlined and the model is validated by polishing experiments on patterned wafers. Finally, practical approaches for minimizing Cu dishing are suggested.

The Influence of Adjacent Segment on the Reliability of Cu Dual Damascene Interconnects

Three terminal ‘dotted-I’ interconnect structures, with vias at both ends and an additional via in the middle, were tested under various test conditions. Mortalities (failures) were found in right segments with jL value as low as 1250 A/cm, and the mortality of a dotted-I segment is dependent on the direction and magnitude of the current in the adjacent segment. Some mortalities were also found in the right segments under a test condition where no failure was expected. Cu extrusion along the delaminated Cu/Si₃N₄ interface near the central via region was believed to cause the unexpected failures. From the time-to-failure (TTF), it is possible to quantify the Cu/Si₃N₄ interfacial strength and bonding energy. Hence, the demonstrated test methodology can be used to investigate the integrity of the Cu dual damascene processes. As conventionally determined critical jL values in two-terminal via-terminated lines cannot be directly applied to interconnects with branched segments, this also serves as a good methodology to identify the critical effective jL values for immortality.