17847039. NON-VOLATILE MEMORY DIE WITH LATCH-BASED MULTIPLY-ACCUMULATE COMPONENTS simplified abstract (Western Digital Technologies, Inc.)
NON-VOLATILE MEMORY DIE WITH LATCH-BASED MULTIPLY-ACCUMULATE COMPONENTS
Organization Name
Western Digital Technologies, Inc.
Inventor(s)
Daniel Joseph Linnen of Naperville IL (US)
Ramanathan Muthiah of Bangalore (IN)
Kirubakaran Periyannan of Saratoga CA (US)
NON-VOLATILE MEMORY DIE WITH LATCH-BASED MULTIPLY-ACCUMULATE COMPONENTS - A simplified explanation of the abstract
This abstract first appeared for US patent application 17847039 titled 'NON-VOLATILE MEMORY DIE WITH LATCH-BASED MULTIPLY-ACCUMULATE COMPONENTS
Simplified Explanation
The abstract of this patent application describes latch-based multiply-accumulate (MAC) operations implemented on the die of a non-volatile memory (NVM) array. The MAC procedures are linear and do not require logic branches. The MAC operation uses a set of linear MAC stages, where each stage processes MAC operations for one bit of a first multi-bit multiplicand multiplied against a second multi-bit multiplicand. The MAC procedures can be used in a neural network feedforward procedure, where the first multiplicand is a synaptic weight and the second multiplicand is an activation value. Multiple plane and multiple die NVM array implementations are also discussed for massive parallel processing.
- Latch-based multiply-accumulate (MAC) operations implemented on the die of a non-volatile memory (NVM) array.
- Linear MAC procedures that do not require logic branches.
- Each linear MAC stage processes MAC operations for one bit of a first multi-bit multiplicand multiplied against a second multi-bit multiplicand.
- MAC procedures can be used in a neural network feedforward procedure.
- First multiplicand is a synaptic weight and the second multiplicand is an activation value.
- Multiple plane and multiple die NVM array implementations for massive parallel processing.
Potential Applications
- Neural networks and deep learning applications.
- High-performance computing and parallel processing.
- Artificial intelligence and machine learning systems.
- Data analytics and pattern recognition.
- Signal processing and image recognition.
Problems Solved
- Efficient implementation of multiply-accumulate (MAC) operations on a non-volatile memory (NVM) array.
- Elimination of logic branches in MAC procedures for improved performance.
- Enabling massive parallel processing for neural networks and other applications.
Benefits
- Improved performance and efficiency in MAC operations.
- Simplified and linear MAC procedures without logic branches.
- Integration of MAC operations on the die of a non-volatile memory (NVM) array.
- Enablement of massive parallel processing for faster computation.
- Potential for lower power consumption and reduced latency.
Original Abstract Submitted
Latch-based multiply-accumulate (MAC) operations implemented on the die of a non-volatile memory (NVM) array are disclosed. The exemplary latch-based MAC procedures described herein are linear procedures that do not require logic branches. In one example, the MAC operation uses a set of linear MAC stages, wherein each linear stage processes MAC operations corresponding to one bit of a first multi-bit multiplicand being multiplied against a second multi-bit multiplicand. Examples are provided wherein the MAC procedures are performed as part of a neural network feedforward procedure where the first multiplicand is a synaptic weight and the second multiplicand is an activation value. Multiple plane and multiple die NVM array implementations are also described for massive parallel processing.