MAROKO133 Breaking ai: New desk-sized ultraviolet light source tool brings China closer to

📌 MAROKO133 Breaking ai: New desk-sized ultraviolet light source tool brings China

Chinese researchers have allegedly developed a new desktop-sized extreme ultraviolet lithography (EUV) light source for producing 14-nanometre microchips. While the new technology cannot replace traditional Advanced Semiconductor Materials Lithography (ASML) machines, it offers an interesting alternative for producing small batches when needed.

According to reports, the new technology could yield other important functions, such as chip inspection, quantum chip prototyping, and photomask defect detection.

The new tech, in other words, is not a mini-ASML EUV machine, but could prove to something incredibly useful for research and development and small-scale fabrication. It is also a significant step toward independence in high-end chipmaking tools.

ASML machines tend to use something called laser-produced plasma (LPP) to blast tin droplets under controlled conditions. Such machines require enormous collector mirrors (which China reportedly cannot yet make domestically).

To the extreme

This new machine, however, reportedly uses a femtosecond laser to fire into argon gas. This setup produces EUV via something called high-harmonic generation.

According to reports, this means the device doesn’t require giant mirrors and doesn’t rely on tin droplets. For this reason, it is a dramatically simpler system and can produce tunable wavelengths from 1 nm to 200 nm.

Traditional ASMLs are also power-hungry things, typically needing around 200 watts of EUV power to expose wafers at production speed. China’s new machine, on the other hand, only consumes around 1 microwatt per blast.

That’s 200 million times less than ASML’s production-level power. This also means that it cannot be used to mass-produce advanced chips. That’s why 14 nm small-batch production is possible, but commercial-scale manufacturing is not.

But (and this is the key point) the “exposure window” is tiny, so brightness per unit area is surprisingly usable for things like inspecting photomasks, checking transistor structures, or etching or exposing tiny experimental chips.

At present, ASML machines are huge things, typically measuring tens of meters in size. These machines are able to produce microchips between 3nm and 7nm in large quantities.

Not a replacement for ASML

If claims are true, this new machine, on the other hand, is much smaller and can produce chips of similar size and quality. It achieves this using high-harmonic generation (HHG) and is small enough to sit on an office desk.

Because of its design and size, the new tech is cheap (only a few percent of a regular ASML machine) and has relatively low running costs. While it can only produce small batch production or lab experiments, it is seen as perfect for inspection and research, not factories.

Despite this, China desperately needs machines like this because it cannot buy ASML’s cutting-edge machines. It also lacks the mirrors needed for ASML-style EUV, and its own EUV programs are lagging globally.

So building an HHG-based EUV source at 13.5 nm (the golden wavelength) is meaningful. According to experts, the new tech is compact, cheap, and good enough for research + small chip batches.

If scaled, it could help Chinese manufacturers improve yields at 14 nm and 28 nm, and it is also a major step in chip inspection and mask alignment. It also embodies China’s plans to reduce its reliance on Western tools in R&D environments while not breaking existing international sanctions.

🔗 Sumber: interestingengineering.com

📌 MAROKO133 Update ai: Humanoid robots get ‘real-world intelligence’, full autonom

OpenMind’s new BrainPack platform brings several core functions of robot autonomy into one modular unit. Instead of separate systems for sensing, mapping, and control, BrainPack combines advanced mapping, object labeling, privacy-protected vision, remote operation, and self-charging in a single backpack-sized device. 

The platform runs on one of Nvidia’s high-performance processing units, giving robots the ability to handle perception and decision-making directly on board rather than relying heavily on external computing.

BrainPack is the same system used in OpenMind’s recent demonstrations of self-charging quadrupeds. Now, the company is making it available to developers, research labs, and early adopters internationally. CEO Jan Liphardt said that the goal is to narrow the gap between robotics and intelligence so that robots can not only move, but also interpret and learn from their surroundings.

Blending research reliability with consumer-level usability 

OpenMind’s CTO Boyuan Chen explains that BrainPack is designed to combine the reliability of a research-grade system with the simplicity and usability of a consumer device. The goal is to make autonomous robotics accessible without needing specialized labs or complex setups. Users can see through the robot’s sensors, guide its learning, and trust that the information it collects is accurate and private.

The platform also acts as the central nervous system for robots, compactly combining high-performance computing, sensors, and software to create a machine that can learn from its surroundings. With it, a robot doesn’t just move – it observes, records, and makes sense of the world around it, while building detailed 3D maps of its environment in real time, reconstructing scenes as it goes. 

The system can recognize and label objects on its own, generating datasets and building a memory of everything it encounters. Privacy is built in, with automatic face detection and blurring to keep humans in view, anonymized. Users can also remotely control the robot and stream video securely from any device, while self-docking charging ensures the robot can operate continuously without interruption.

Robots map, patrol, and self-charge 

BrainPack allows robots to gain advanced capabilities through simple plug-and-play integration, making them able to operate in homes, research labs, and public spaces. Early testing by OpenMind has shown that BrainPack-equipped robots can perform self-guided patrols, map multi-room environments, recognize and label objects, and return to charging stations automatically, all without direct supervision. 

The platform demonstrates a practical way to give robots useful autonomy while keeping setup and operation straightforward. Earlier this year, the startup introduced FABRIC, a protocol that lets robots verify identity and share context with each other. The system enables machines to learn from one another in real time, adapt to new environments, and communicate more effectively, making collaboration between humans and robots smoother and more intuitive.

Furthermore, OpenMind has developed OM1, a hardware-agnostic operating system for humanoid robots. OM1 is built from the ground up as an AI-native platform, capable of running on a wide range of robotic hardware and acting as a universal brain for machines. 

🔗 Sumber: interestingengineering.com