Chip design is the key and the most challenging part of chip production. The process includes: selecting the appropriate materials and structures, determining the proper design methods and flow, and designing systems with specific functional and performance requirements. In the process of physical chip design, after completing timing and functional convergence, it enters the DFM stage, when the layout, whether it is the density of the metal or the width of the graph, is not uniform; since the flatness of the wafer surface cannot be guaranteed, the redundant filling must be used to improve the flatness of the chip. A redundant fill is a layout structure that does not affect the logic function of the chip. A redundant filling helps to improve the consistency of interconnect pattern density and is a crucial method to optimize the CMP process.
WiMi Hologram Cloud, Inc. (NASDAQ:WIMI) developed an artificial intelligence-based chip design redundancy automated fill system. Chip design redundancy filling usually uses rule-based and pane-based redundancy filling methods.
The rule-based redundant metal filling algorithm performs a redundant metal filling using two or more steps. Better image gradients are achieved by filling redundant shapes with larger sizes and buffer distances and serving smaller ones. The use of multi-step filling and the introduction of redundant graphics with smaller sizes and buffer distances lead to further improvements in the overall image density.
The redundant metal filling method based on the window format results in better graphic density and graphic gradients. The technique uses a specific algorithm to fill the redundant metals according to the surroundings of the design layout. First, the whole chip is divided into individual panes (e.g., 50 μm), then the graphic density in each pane is calculated, and finally, the amount of redundant metal filling is decided based on the expressive thickness of adjacent panes. The goal of the repetitious metal filling is to minimize the graphic gradient while the window format fills through a circular algorithm to make the minimum amount of filled redundant metal, while the density gradient difference between each adjacent pane is also minimized.
Both filling methods are under-constrained. The former filling method does not consider the graphic gradient, which is an important reason for the generation of hot spots in CMP and etching processes, and the latter leads to the existence of redundant inconsistency in the metal filling, which affects the uniformity of etching and CMP processes.
WiMi’s R&D team uses artificial intelligence technology for chip redundancy automatic filling system on this basis, which selects and calculates filling density and gradient through artificial intelligence algorithm technology. The artificial intelligence chip design redundant automated filling system obtains the redundant graphic filling template map with a high yield rate, corresponding to the chip layout that needs to be filled as the training image, and obtains the filling position label and gradient of the redundant graphic through artificial intelligence model calculation; chip design redundant filling allows the chip production process to maintain the consistency of interconnection graphic density, which significantly improves the production yield of chips. In particular, the production yield of large-capacity memory is generally low, making the chip yield rate lower. By using chip design redundancy filling, it makes it possible to place multiple memories inside the chip, and even if one or more of them have problems and fail the test, as long as the remaining correct memories in the chip meet certain conditions, through calibration and specific configuration, the whole chip will not be affected, which can improve the chip The production yield of the chip can be improved.