Opening the new AI Era with Analog Computing Hardware.
IHW is creating true AI (Artificial Intelligence) chips with analog computing capabilities, so-called APU (Analog Processing Unit) to overcome the limits of modern digital computing technologies.
Our APUs will surpass current hardware units, which are the GPU (Graphic Processing Unit), TPU (Tensor Processing Unit), and NPU (Neural Processing Unit) by providing a whole new level of power efficiency and comparable inference accuracy.
This technology will enable on-device inferencing for edge computing and mobile devices and ultimately replace the current GPU and TPU for inferencing.
IHW is at the forefront of attaining viable analog computing technologies, which is currently known to only be possible at a rudimentary level, but not at a commercially-feasible level.
Our APUs will not only demonstrate true analog computing that exceed expectations but will have inferencing accuracies comparable to digital inferencing.
These achievements will be made possible by applying new hardware (HW) technologies to memory devices and circuit designs in combination with optimized algorithms.
Global network of IHW
Why is the APU so powerful?
New Level of Power-Saving
- no memory wall
Incomparable Parallelism and Efficiency
- All-in-one MAC Calculation
Significant Cost Reduction
- from GPU or TPU
Digital-Comparable Accuracy
Evolution of computing architecture
Modern software-driven AI is facilitated by GPU & TPU for parallel computing, but its capabilities are restricted by excess power consumption, large area occupation, and higher cost that will limit its applications.
Superiority of APU
1. Ultra-fast MAC operation
Digital Matrix Calculation
✓ If N x N matrix
→ MULTIPLES=N x N2=N3
→ ADDS= (N-1) x N2
→ TOTAL 2N3 – N2 calculation
Analog Matrix Calculation
source : https://youtu.be/euR76oRd7L4
✓ Regardless of the matrix size, calculation can be done simultaneously.
2. Super power saving (no memory wall)
For ideal AI, removing memory wall is inevitable to enhance the computing efficiency and to reduce the power consumption
3. Overwhelming Energy Efficiency (TOPs/Watt)
Future Impacts and Applications
Mobile Devices &
Edge Computing
Mobile and edge computing
Autonomous Cars
Medical Diagnosis
High accuracy Medical application
Robots & Drones
Stock Analysis
& Banking
Stock analysis & banking
Generative AI
Initial target Application :
Mobile Devices & Edge Computing
Due to the low power consumption, low costs, and small area occupation, APUs can replace GPU/TPUs and create new mobile H/W markets such as image processing, voice processing, security & surveillance, consumer commercial drone, objective recognition, autonomous cars.
Image Processing
Voice Processing
Commercial Drones
Autonomous Car
Pattern Recognition
Security & Surveillance
Model-F
Inference focused
Model-A
Inference and On-Chip Training
Model-I
Higher Density and Lower Cost
Available Markets for each model
1. Replacing GPU/TPU with cloud/data centers for training
2. Replacing CPU/GPU/ASIC/FPGA for inferencing
3. Entering the newly growing H/W market for edge/terminal computing
✓ Enabled by energy efficient and fast analog computing