Research on the Evolution and Classification of Artificial Intelligence Chip Technology: A Review of Architecture Characteristics, Algorithm Adaptation, and Application Scenarios

Yinbo Hao

doi:10.54097/4sdjvb71

Authors

Yinbo Hao

DOI:

https://doi.org/10.54097/4sdjvb71

Keywords:

Artificial intelligence chips, technological evolution, chip classification, architectural characteristics, algorithm adaptation, application scenarios

Abstract

This paper systematically reviews the technological evolution, classification systems, and application scenarios of AI chips. Research shows AI chips have evolved from traditional general - purpose processors to a diverse ecosystem including GPUs, FPGAs, ASICs (e.g., TPUs, NPUs), and brain - inspired chips. By analyzing architectural features, algorithm adaptation, and performance limits of various chips, a multidimensional classification framework is constructed, categorizing AI chips from three dimensions: technical architecture, functional positioning, and application scenarios. The study finds general - purpose chips (e.g., GPUs) are suitable for cloud training but have low energy efficiency, specialized chips (e.g., TPUs) have excellent energy efficiency in inference but lack flexibility, and edge computing chips balance power consumption and performance. Future AI chip development will trend towards heterogeneous integration, memory - compute integration, and hardware - software co - design to tackle challenges like computational efficiency, memory wall issues, and algorithm diversity. This research offers theoretical guidance and practical references for chip selection, architectural optimization, and application deployment.

Downloads

Download data is not yet available.

References

[1] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436-444.

[2] Jouppi, N. P., Young, C., Patil, N., Patterson, D., Agrawal, G., Bajwa, R., ... & Boyle, R. (2017). In-datacenter performance analysis of a tensor processing unit. Proceedings of the 44th Annual International Symposium on Computer Architecture, 1-12.

[3] Moore, G. E. (1965). Cramming more components onto integrated circuits. Electronics, 38(8), 114-117.

[4] Coates, A., Huval, B., Wang, T., Wu, D., Catanzaro, B., & Andrew, N. (2013). Deep learning with COTS HPC systems. International Conference on Machine Learning, 1337-1345.

[5] Sze, V., Chen, Y. H., Yang, T. J., & Emer, J. S. (2017). Efficient processing of deep neural networks: A tutorial and survey. Proceedings of the IEEE, 105(12), 2295-2329.

[6] Hennessy, J. L., & Patterson, D. A. (2019). A new golden age for computer architecture. Communications of the ACM, 62(2), 48-60.

[7] Chen, Y. H., Emer, J., & Sze, V. (2016). Eyeriss: A spatial architecture for energy-efficient dataflow for convolutional neural networks. ACM SIGARCH Computer Architecture News, 44(3), 367-379.

[8] Wang, X., Han, Y., Leung, V. C., Niyato, D., Yan, X., & Chen, X. (2020). Convergence of edge computing and deep learning: A comprehensive survey. IEEE Communications Surveys & Tutorials, 22(2), 869-904.

[9] Sze, V., Chen, Y. H., Yang, T. J., & Emer, J. S. (2017). Efficient processing of deep neural networks: A tutorial and survey. Proceedings of the IEEE, 105(12), 2295-2329.