US Patent Application 18216797. INTEGRATED CIRCUITS WITH MACHINE LEARNING EXTENSIONS simplified abstract
Contents
INTEGRATED CIRCUITS WITH MACHINE LEARNING EXTENSIONS
Organization Name
Inventor(s)
Martin Langhammer of Alderbury (GB)
Dongdong Chen of Union City CA (US)
Kevin Hurd of Santa Clara CA (US)
INTEGRATED CIRCUITS WITH MACHINE LEARNING EXTENSIONS - A simplified explanation of the abstract
- This abstract for appeared for US patent application number 18216797 Titled 'INTEGRATED CIRCUITS WITH MACHINE LEARNING EXTENSIONS'
Simplified Explanation
This abstract describes an integrated circuit that has specialized processing blocks optimized for machine learning algorithms. These processing blocks have a multiplier data path that is divided into multiple partial product generators, compressors, and carry-propagate adders. The results from the carry-propagate adders are then added using a floating-point adder. Optionally, the results can be cast to a higher precision. The adder data path includes an adder that combines the results from the floating-point adder with zero, a general-purpose input, or other dot product terms. This specialized processing block greatly increases the functional density for implementing machine learning algorithms.
Original Abstract Submitted
An integrated circuit with specialized processing blocks is provided. A specialized processing block may be optimized for machine learning algorithms and may include a multiplier data path that feeds an adder data path. The multiplier data path may be decomposed into multiple partial product generators, multiple compressors, and multiple carry-propagate adders of a first precision. Results from the carry-propagate adders may be added using a floating-point adder of the first precision. Results from the floating-point adder may be optionally cast to a second precision that is higher or more accurate than the first precision. The adder data path may include an adder of the second precision that combines the results from the floating-point adder with zero, with a general-purpose input, or with other dot product terms. Operated in this way, the specialized processing block provides a technical improvement of greatly increasing the functional density for implementing machine learning algorithms.
