What is an NPU?

29 November 2024 AI & ML

An NPU stands for Neural Processing Unit, which is a specialised hardware accelerator designed to efficiently process tasks related to artificial intelligence, in particular deep learning models. NPUs are optimised for operations commonly found in neural networks, such as matrix multiplications, convolutions, and activation functions.

Here are some key points about NPUs:

NPUs are designed to perform the following AI tasks at maximum speed, while using the least amount of energy. NPUs are:

• Optimised for AI workloads: NPUs are designed to handle the vast number of computations involved in neural network processing more efficiently than general-purpose processors (like CPUs or GPUs).

• Power efficient: They are typically more power-efficient than using CPUs or even GPUs for AI tasks, making them ideal for mobile devices and edge computing where power consumption is critical.

• Able to perform real-time processing: NPUs can execute AI tasks, such as image recognition, speech processing, and other inference tasks, in real-time due to their high computational parallelism.

• Used in smartphones and edge devices: NPUs are increasingly common in modern smartphones (like those made by Apple, Huawei, and Google) to accelerate AI tasks such as photography enhancements, face recognition, and voice assistance.

• Integrated with SoCs (System on Chips): Many NPUs are embedded within SoCs, which are used in devices like smartphones, autonomous vehicles, IoT devices, and more, enabling AI computations directly on the device without needing to rely on cloud-based processing.

To summarise, neural processing units are crucial for enabling the growing demand for AI processing in compact, energy-efficient, and real-time environments without needing to be connected to a server to handle the AI processing.

Further reading:

Compact direct Time-of-Flight 3D LiDAR module
Altron Arrow AI & ML DSP, Micros & Memory
The VL53L9 from STMicroelectronics is the first direct Time-of-Flight (dToF) 3D LiDAR all-in-one module in ST’s portfolio, offering a resolution of 2,3K zones, wide field of view, on-chip processing, 100 frames per second, and sensing range from 5 centimeters to 9 meters.

Read more...

---

SoC brings scalable edge AI to life
Altron Arrow AI & ML
The NXP i.MX 937 applications processor is optimised for performance without excessive power draw and bridges the gap between entry-level chips and high-end processors.

Read more...

---

Wall-mount cooling for Edge IT
AI & ML
Designed to run 24/7 in critical compute environments, Vertiv CoolPhase Wall delivers up to 60% greater airflow than comfort cooling.

Read more...

---

Compact direct Time-of-Flight 3D LiDAR module
Altron Arrow Opto-Electronics Electronics Technology AI & ML
The VL53L9 from STMicroelectronics is the first direct Time-of-Flight (dToF) 3D LiDAR all-in-one module in ST’s portfolio, offering a resolution of 2,3K zones, wide field of view, on-chip processing, 100 frames per second, and sensing range from 5 centimeters to 9 meters.

Read more...

---

Edge AI improves condition monitoring
EBV Electrolink AI & ML
STMicroelectronics has introduced the IIS3DWB10IS, an intelligent MEMS vibration sensor designed for high-performance industrial condition monitoring applications.

Read more...

---

Tiny SoM integrates NXP i.MX 91
AI & ML
The QS91 module from Direct Insight utilises the low-cost single core i.MX 91 to deliver secure, energy-efficient Linux capabilities for edge applications.

Read more...

---

Robotics platform powers intelligent systems
Vepac Electronics AI & ML
The CEXD-INTRBL open robotics development system combines high-performance AI processing, sensor integration, and motion control into a compact embedded platform.

Read more...

---

SensorGPT speeds intelligent IoT deployment
AI & ML
TDK has introduced SensorGPT, a new technology designed to accelerate the deployment of intelligent edge AI and smart IoT applications by dramatically reducing reliance on real-world sensor data.

Read more...

---

Ideal entry point into edge AI
Avnet Silica AI & ML
congatec has released new COM Express modules with Intel Core Series 3 processors for cost- and energy-optimised embedded computing applications.

Read more...

---

Analogue reservoir AI chip capable of real-time learning
Altron Arrow AI & ML
TDK Corporation has jointly developed a prototype of a reservoir AI chip using an analogue electronic circuit that mimics the cerebellum with Hokkaido University.

Read more...

---