e-AI Capability Advancements
■
Renesas is evolving e-AI. Classes 1 through 4, and beyond,
increase capability incrementally at each step while keeping
similar power consumption
■
Exclusive Dynamically Reconfigurable Processor (DRP)
technology and architecture accelerate image processing,
object recognition, AI, and cognitive decision making
e-AI Capability
2018 2019 2020 2021 2022 2017
Endpoint Inference
e-AI on MCU/MPU
Class-1
X10
Real-time Image Processing
by DRP
Real-time Cognition
by DRP- AI
Endpoint Incremental Learning
by DRP- AI2
Class-2
Class-3
Class-4
X10
X10
Next
Next
Solution released July 2017
RZ/A2M product release October 2018
Paper reported VLSI Symp. 2018
Learning in Cloud
AI Inference
at Endpoint
Pre- or
One-time
Training
Learning & Inference
in Cloud
Statistical Application
OT
Operational Technology
IT
Information Technology
Endpoint
Cloud
Edge
Real-time action
No Cloud Lag
Real-time Application
e-AI: Local Real-time AI by Inference
■
Traditional statistical AI applications execute completely
in the Cloud
■
Real-time applications cannot tolerate cloud lag at the
endpoint
■
e-AI takes immediate action locally through inference
from cloud-trained AI neural networks
■
Each evolution step represents 10 times the previous computing
power due to DRP (see below) advancement
■
Class 4 represents capability of incremental learning without
connection to the Cloud to solve complex graphical problems and
process multi-sensor inputs for robotics