Grey Neural Network Based Predictive Model For Multi-core Architecture Spatial Characteristics

Early design space exploration is an essential ingredient in modern processor development. As the number of cores on a processor increases, these large and sophisticated multi-core-oriented architectures exhibit increasingly complex and heterogeneous characteristics. In the upcoming multi-/many-core era, the design, evaluation and optimization of architectures will demand analysis methods that are very different from those targeting traditional, centralized and monolithic hardware structures. A common weakness of existing analytical models lack the ability to forecast the complex and heterogeneous behavior of large and distributed architecture substrates across the design space. This limitation will only be exacerbated with the rapidly increasing integration scale. Now grey neural network predictive models and 2D wavelet transform techniques is applied, which can efficiently reason the characteristics of large and sophisticated multi-core oriented architectures during the design space exploration stage without using detailed cycle-level simulations.Grey system theory and neural network are two of popular data predicting methods. By means of studying the two methods, we find that there is certain similarity in information expression, meanwhile existing differences and complementarities. Therefore, combining grey system with neural network, the grey neural network can make up the shortage of using single model to achieving excellent data processing and predictive validity.In this paper, a novel grey neural network model is proposed based on grey system, neural network and their combing patterns. Combing the improved GM(0,N) with improved RBF neural network, the SGRBF model has the feature of not only grey model, which applies to small sample predicting and don't require special data distribution, but also neural network, which has fine nonlinear approximate ability, fast learning speed and parallel processing ability. Complementing each other's advantages to solving small sample data predicting problem better.The grey neural network model is applied to design space exploration. And grey neural network is mainly employed to predict 2D space parameters produced by wavelet decomposition for achieving better predictive accuracy with less samples. The experimental result shows that grey neural network can achieve high predictive accuracy, better stability and reliability, the whole model can efficiently reason the characteristics of large and sophisticated multi-core oriented architectures during the design space exploration stage. In addition, using grey neural network to multi-core architecture spatial exploration, can broad the range of its application field, provide a novel and effective method to multi-core processor design meanwhile, which is important and significant for current research.