TingTing Hwang

TSV redundancy: architecture and design issues in 3D IC

3D technology provides many benefits including high density, high band-with, low-power, and small form-factor. Through Silicon Via (TSV), which provides communication links for dies in vertical direction, is a critical design issue in 3D integration. Just like other components, the fabrication and bonding of TSVs can fail. A failed TSV may cause a number of known-good-dies that are stacked together to be discarded. This can severely increase the cost and decrease the yield as the number of dies to be stacked increases. A redundant TSV architecture with reasonable cost for ASICs is proposed in this paper. Design issues including recovery rate and timing problem are addressed. Based on probabilistic models, some interesting findings are reported. First, the probability that three or more TSVs are failed in a tier is less than 0.002%. Assumption of that there are at most two failed TSVs in a tier is sufficient to cover 99.998% of all possible faulty free and faulty cases. Next, with one redundant TSV allocated to one TSV block, limiting the number of TSVs in each TSV block to be no greater than 50 and 25 leads to 90% and 95% recovery rates when 2 failed TSVs are assumed. Finally, analysis on overall yield shows that the proposed design can successfully recover most of the failed chips and increase the yield of TSV bonding to 99.99%. This can effectively reduce the cost of manufacturing 3D ICs.

Compiler optimization on instruction scheduling for low power

In this paper, we investigate the compiler transformation techniques to the problem of scheduling VLIW instructions aimed to reduce the power consumption on the instruction bus. It can be categorized into two types: horizontal and vertical scheduling. For the horizontal case, we propose a bipartite-matching scheme. We prove that our greedy algorithm always gives the optimal switching activities of the instruction bus. In the vertical case, we prove that the problem is NP-hard, and propose a heuristic algorithm. Experimental results show average 13% improvements with 4-way issue architecture and average 20% improvement with 8-way issue architecture for power consumptions of instruction bus as compared with conventional list scheduling for an extensive set of benchmarks.

Low power realization of finite state machines—a decomposition approach

We present in this article a new approach to the synthesis problem for finite state machines with the reduction of power dissipation as a design objective. A finite state machine is decomposed into a number of coupled submachines. Most of the time, only one of the submachines will be activated which, consequently, could lead to substantial savings in power consumption. The key steps in our approach are: (1) decomposition of a finite state machine into submachines so that there is a high probability that state transitions will be confined to the smaller of the submachines most of the time, and (2) synthesis of the coupled submachines to optimize the logic circuits. Experimental results confirmed that our approach produced very good results (in particular, for finite state machines with a large number of states.)

TSV Redundancy: Architecture and Design Issues in 3-D IC

3-D technology provides many benefits including high density, high bandwidth, low-power, and small form-factor. Through Silicon Via (TSV), which provides communication links for dies in vertical direction, is a critical design issue in 3-D integration. Just like other components, the fabrication and bonding of TSVs can fail. A failed TSV can severely increase the cost and decrease the yield as the number of dies to be stacked increases. A redundant TSV architecture with reasonable cost is proposed in this paper. Based on probabilistic models, some interesting findings are reported. First, the number of failed TSVs in a tier is usually less than 2 when the number of TSVs in a tier is less than 1000 and less than 5 when the number of TSVs in a tier is less than 10000. Assuming that there are at most 2-5 failed TSVs in a tier. With one redundant TSV allocated to one TSV block, our proposed structure leads to 90% and 95% recovery rates for TSV blocks of size 50 and 25, respectively. Finally, analysis on overall yield shows that the proposed design can successfully recover most of the failed chips and increase the yield of TSV to 99.4%.

Through-Silicon Via Planning in 3-D Floorplanning

In this paper, we will study floorplanning in 3-D integrated circuits (3D-ICs). Although literature is abundant on 3D-IC floorplanning, none of them consider the areas and positions of signal through-silicon vias (TSVs). In previous research, signal TSVs are viewed as points during the floorplanning stage. Ignoring the areas, positions and connections of signal TSVs, previous research estimates wirelength by measuring the half-perimeter wirelength of pins in a net only. Experimental results reveal that 29.7% of nets possess signal TSVs that cannot be put into the white space within the bounding boxes of pins. Moreover, the total wirelength is underestimated by 26.8% without considering the positions of signal TSVs. The considerable error in wirelength estimation severely degrades the optimality of the floorplan result. Therefore, in this paper, we will propose a two-stage 3-D fixed-outline floorplaning algorithm. Stage one simultaneously plans hard macros and TSV-blocks for wirelength reduction. Stage two improves the wirelength by reassigning signal TSVs. Experimental results show that stage one outperforms a post-processing TSV planning algorithm in successful rate by 57%. Compared to the post-processing TSV planning algorithm, the average wirelength of our result is shorter by 22.3%. In addition, stage two further reduces the wirelength by 3.45% without any area overhead.

Randomness Enhancement Using Digitalized Modified Logistic Map

In this brief, a nonlinear digitalized modified logistic map-based pseudorandom number generator (DMLM-PRNG) is proposed for randomness enhancement. Two techniques, i.e., constant parameter selection and output sequence scrambling, are employed to reduce the computation cost without sacrificing the complexity of the output sequence. Statistical test results show that with only one multiplication, DMLM-PRNG passes all cases in SP800-22. Moreover, it passes most of the cases in Crush, one of the test suites of TesuU01. When compared with solutions based on digitized pseudochaotic maps previously proposed in the literature, in terms of randomness quality, our system is as good as a Rényi-map-based PRNG and better than a logistic-map-based PRNG. Moreover, compared with solutions based on a Rényi-map-based PRNG, DMLM-PRNG is better scalable to high digital resolutions with reasonable area overhead.

Compiler optimization on VLIW instruction scheduling for low power

In this article, we investigate compiler transformation techniques regarding the problem of scheduling VLIW instructions aimed at reducing power consumption of VLIW architectures in the instruction bus. The problem can be categorized into two types: horizontal scheduling and vertical scheduling. For the case of horizontal scheduling, we propose a bipartite-matching scheme for instruction scheduling. We prove that our greedy bipartite-matching scheme always gives the optimal switching activities of the instruction bus for given VLIW instruction scheduling policies. For the case of vertical scheduling, we prove that the problem is NP-hard, and we further propose a heuristic algorithm to solve the problem. Our experiment is performed on Alpha-based VLIW architectures and an ATOM simulator, and the compiler incorporated in our proposed schemes is implemented based on SUIF and MachSUIF. Experimental results of horizontal scheduling optimization show an average 13.30% reduction with four-way issue architecture and an average 20.15% reduction with eight-way issue architecture for transitional activities of the instruction bus as compared with conventional list scheduling for an extensive set of benchmarks. The additional reduction for transitional activities of the instruction bus from horizontal to vertical scheduling with window size four is around 4.57 to 10.42%, and the average is 7.66%. Similarly, the additional reduction with window size eight is from 6.99 to 15.25%, and the average is 10.55%.

ELM-a fast addition algorithm discovered by a program

A new addition algorithm, ELM, is presented. This algorithm makes use of a tree of simple processors and requires O(log n) time, where n is the number of bits in the augend and addend. The sum itself is computed in one pass through the tree. This algorithm was discovered by a VLSI CAD tool, FACTOR, developed for use in synthesizing CMOS VLSI circuits. >

Power reduction by gate sizing with path-oriented slack calculation

This paper describes methods for reducing power consumption. We propose using gate sizing technique to reduce power for circuits that have already satisfied the timing constraint. Replacement of gates on noncritical paths with smaller templates is used in reducing the dissipated power of a circuit. We find that not only gates on noncritical paths can be down-sized, but also gates on critical paths can be down-sized. A power reduction algorithm by means of single gate resizing as well as multiple gates resizing will be proposed. In addition, to identify gates to be resized, a path-oriented method in calculating slack time with false path taken into consideration will be also proposed. During the slack time computation, in order to prevent long false path from becoming sensitizable and thus increasing the circuit delay, slack constraint will be set for gales. Results on a set of circuits from MCNC benchmark set demonstrate that our power reduction algorithm can reduce about 10% more power, on the average, than a previously proposed gate sizing algorithm.

Combining technology mapping and placement for delay-optimization in FPGA designs

We combine technology mapping and placement into a single procedure, M.Map, for the design of RAM-based FPGAs. Iteratively, M.Map maps several subnetworks of a Boolean network into a number of CLBs on the layout plane simultaneously. For every output node of the unmapped portion of the Boolean network, many ways of mapping are possible. The choice of which mapping to be used depends not only on the location of the CLB into which the output node will be mapped but also on its interconnection with those already mapped CLBs. To deal with such a complicated interaction among multiple output nodes of a Boolean network, multiple ways of mappings and multiple number of CLBs, any greedy algorithm will be insufficient. Therefore, we use a bipartite weighted matching algorithm in finding a solution that takes the global information into consideration. With the availability of the partial placement information, M.Map is able to minimize the routing delay in addition to the number of CLBs. Experimental results on a set of benchmarks demonstrate that M.Map is indeed effective and efficient. >