Current-Mode MAC Unit for Linear Low-Power Signal Accumulation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Hardware systems performing MAC operations face challenges with non-linear conversions due to voltage mode DACs and voltage-to-current conversions, leading to reduced dynamic range and increased power consumption, along with the need for additional filters and active stages.

Innovation Solution

The implementation of a mixed mode analog-digital architecture using current mode devices as filters, which eliminates the need for active filters and drivers, enhancing linearity and reducing device size and power consumption, while providing a direct interface with non-volatile memory types like MRAM.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If voltage mode DACs are used to perform MAC operations, then the system can perform multiplication and addition operations, but non-linear conversions occur leading to reduced dynamic range

Engineering Contradiction:
Improvedynamic rangeVSAvoidlinearity
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent replaces voltage mode DACs with current mode DACs. Current mode operation eliminates the non-linear voltage-to-current conversion stage, providing inherently linear operation. The current mode DACs directly generate current outputs that can be summed without conversion losses, thereby maintaining both high dynamic range and linearity simultaneously.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If voltage mode DACs with filters and active stages are used, then MAC operations can be performed, but power consumption increases

Engineering Contradiction:
ImproveMAC operation capabilityVSAvoidpower consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent extracts and removes the unnecessary voltage mode conversion stages and associated active filters from the system. By using current mode DACs that directly output currents, the design eliminates power-hungry voltage buffers and conversion circuits, achieving MAC operations with significantly reduced power consumption.

Inventive Principle:
Principle #2Taking out (Extraction)

3Productivity

If voltage mode DACs with additional filters and active stages are used, then MAC operations can be performed, but device size increases

Engineering Contradiction:
ImproveMAC operation capabilityVSAvoiddevice size
Core Design Contradiction:
ProductivityVSArea of stationary object

Solution Approach 1:

The patent removes the additional active stages and filter circuits that are required in voltage mode implementations. Current mode DACs integrate the multiplication and addition functions directly in the current summation node, eliminating the need for separate voltage buffers, active filters, and conversion stages, thereby reducing overall device area.

Inventive Principle:
Principle #2Taking out (Extraction)

4Reliability

If current mode DACs are used to multiply input signal with current, then linearity is enhanced, but the system requires direct interface with non-volatile memory

Engineering Contradiction:
ImprovelinearityVSAvoidinterface compatibility
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent changes the operational parameter domain from voltage mode to current mode throughout the MAC unit. Current mode operation provides inherent linearity and naturally interfaces with non-volatile memory devices like MRAM that operate in the current domain, eliminating the need for voltage-to-current conversion stages and maintaining both linearity and interface compatibility.

Inventive Principle:
Principle #35Parameter changes

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach improves linearity and reduces power consumption and device size by eliminating the need for filters and drivers, while enabling efficient operations with non-volatile memory and improved performance in MAC operations.

Implementation Method 1

a first current mode digital-to-analog converter (DAC) configured to multiply an input signal with a first current having a first amplitude to generate a first signal

Methodology Applied
Scientific EffectCurrent mode multiplication:

Implementation Method 2

a second current mode DAC configured to multiply the input signal with a second current having a second amplitude to generate a second signal

Methodology Applied
Scientific EffectCurrent mode multiplication:

Implementation Method 3

a mixer configured to multiply the second signal with a clock signal to generate a third signal

Methodology Applied
Scientific EffectSignal mixing:

Implementation Method 4

The third signal can be combined with the first signal via a current summing node to generate an output signal

Methodology Applied
Scientific EffectCurrent summation:

Data Source

PatentUS11379186B2Mixed mode multiply and accumulate unit
Publication Date: 2022.07.05 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • US11379186B2 patent drawing
  • US11379186B2 patent drawing
  • US11379186B2 patent drawing

AI summary

Systems and methods to implement a multiply and accumulate (MAC) unit is described. In an example, a device can include a first current mode digital-to-analog converter (DAC) configured to multiply an input signal with a first current having a first amplitude to generate a first signal. The device can further include a second current mode DAC configured to multiply the input signal with a second current having a second amplitude to generate a second signal. The second amplitude can be less than the first amplitude. The device can further include a mixer configured to multiply the second signal with a clock signal to generate a third signal. The third signal can be combined with the first signal via a current summing node to generate an output signal. The output signal can be outputted to another device.