The U.S. Army has awarded defense and technology company Palladyne AI two contracts to test its AI-powered swarm software and reusable strike drone with soldiers during field exercises in Colorado and California.

The contracts, awarded under the Army’s Disruptive Applications Broad Agency Announcement, will support operational validation of the company’s SwarmOS autonomy platform and the Gremlin-X unmanned aerial vehicle.

The technologies will be evaluated by soldiers from the Army’s 4th Infantry Division during reconnaissance and target acquisition exercises. Palladyne AI said the demonstrations will use Army-specified platforms and the Android Team Awareness Kit (ATAK), a command-and-control system widely used by U.S. forces.

The effort reflects the Army’s growing interest in autonomous systems that can operate in environments where communications are limited or disrupted.

One operator, many drones

SwarmOS is an autonomy software stack designed to allow a single operator to control multiple autonomous systems from different manufacturers at the same time.

According to the Utah-based company, the software runs on edge hardware and is designed to function in degraded or denied communications environments. Rather than relying on centralized control infrastructure, the platform enables autonomous systems to coordinate with one another while operating in the field.

“The hardest problem in autonomous systems isn’t making one platform smart, it’s making heterogeneous platforms from different manufacturers collaborate intelligently without dependence on centralized infrastructure that adversaries can target and defeat,” said Dr. Denis Garagic, Chief Technology Officer and Principal Investigator, Palladyne AI.

He added: “We solved that.”

The Army exercises will evaluate how the software supports reconnaissance and target acquisition missions while integrating with existing military command-and-control tools.

Reusable strike drone

The second technology being tested is Gremlin-X, a Group 2 multirotor drone being developed as a reusable loitering munition.

Unlike traditional loitering munitions that are typically lost after striking a target, Gremlin-X is designed to return and be recovered after a mission. The company said this approach could reduce costs because only the munition payload would need replacement rather than the entire aircraft.

Gremlin-X is intended to operate in denied, degraded, intermittent, and limited communications environments, often referred to as DDIL conditions. The drone is being developed to support small, dispersed military formations that may need to engage targets without relying on centralized command networks.

Beyond the 4th Infantry Division exercises, Palladyne AI plans to demonstrate SwarmOS, IntelliSwarm, and Gremlin-X during Northern Strike 26-2, a large U.S. military training exercise involving more than 9,000 participants.

“This contract marks a decisive shift in how the U.S. Army is engaging with autonomous systems, moving from evaluation to funded operational use in the field,” said Doug Dynes, President of Palladyne Aerospace and Defense.

The company said the demonstrations will help validate the technologies in operational settings and could support future defense opportunities as the military continues exploring autonomous systems, drone swarms, and human-machine teaming concepts for battlefield operations.

European Space Agency (ESA) engineers are assembling one of Europe’s most advanced robotic arms, a system designed to handle samples and perform complex tasks during future missions to the Moon and Mars.

Known as the Sample Transfer Arm (STA), the robot can extend up to 2.4 meters and move with seven degrees of freedom, giving it a range of motion similar to a human arm. The system is equipped with cameras, sensors, and onboard electronics that allow it to perceive its surroundings and perform tasks with a high level of autonomy.

The robotic arm was originally developed as part of the joint NASA-ESA Mars Sample Return campaign. Its role was to transfer Martian samples collected by NASA’s Perseverance rover for eventual return to Earth.

Although the future of Mars Sample Return remains uncertain, the technology is now being positioned for a wider range of applications, including future lunar exploration missions.

Human-like space helper

The arm features a shoulder, elbow, and wrist configuration similar to a human arm. At its tip is a gripper capable of handling objects with millimeter-level precision.

The system also includes a sophisticated force and torque sensor that acts as a sense of touch. The sensor measures how an object is being pushed, pulled, or twisted in three-dimensional space, allowing the arm to manipulate items with greater accuracy.

Position sensors embedded throughout the joints continuously calculate the location of the arm’s tip, helping maintain precise control during operations.

Engineers say this level of precision could prove useful for a variety of activities in space, including collecting geological samples, transferring equipment, and supporting astronauts during surface missions.

The robotic arm also contains its own electronics unit, which acts as the system’s control center. Cameras mounted on the arm provide visual awareness, enabling the robot to detect and interact with objects in its environment.

Ready for testing

Teams at Italian aerospace company Leonardo are currently completing the integration of the system at the company’s facility in Nerviano, near Milan.

In the coming weeks, engineers will begin testing the arm in simulated space environments to evaluate its performance before any future deployment.

The project brings together companies and suppliers from across Europe. The industrial consortium is led by Leonardo and includes GMV and AVS from Spain, Maxon from Switzerland, 3DPlus from France, and COMOTI from Romania, along with suppliers from Denmark, Greece, and Germany.

The arm incorporates several specialized systems designed for spaceflight. These include a flexible wrist mechanism that allows softer contact with objects, a flat electrical harness that carries hundreds of signals throughout the arm, and structural mechanisms that help withstand the forces experienced during launch and landing.

European space agencies view advanced robotics as a key capability for future exploration missions, particularly as crews prepare for longer stays on the Moon and eventual missions to Mars.

With its ability to see, feel, and autonomously manipulate objects, the Sample Transfer Arm could become an important tool for handling samples and supporting scientific operations beyond Earth.

Researchers at Harvard University have developed a semiconductor chip that can synthesize 64 different DNA sequences in parallel using electric currents and a water-based enzymatic process, potentially offering an alternative to conventional DNA manufacturing methods.

The chip uses localized electrical control to trigger DNA synthesis at selected sites on its surface. The team says the approach avoids the solvent-heavy phosphoramidite chemistry widely used to produce synthetic DNA today.

Synthetic DNA is a key tool in modern biotechnology, supporting applications ranging from diagnostics and genome engineering to cancer research. While enzymatic DNA synthesis has emerged as a more environmentally friendly alternative, it has struggled to match the scale of conventional methods.

According to the researchers, enzymatic approaches have so far been limited to producing about a dozen DNA sequences simultaneously. In the new study, the chip generated 64 distinct DNA sequences in parallel, with each sequence reaching up to 39 nucleotides in length.

Electricity guides DNA

DNA synthesis occurs one nucleotide at a time. After each addition, a temporary blocking group must be removed before the next nucleotide can be attached.

To control this process, the Harvard team designed a chip with 64 synthesis sites. Each site contains two concentric ring electrodes surrounding DNA strands anchored at the center.

When activated, the inner electrode generates protons that lower the pH around a selected DNA strand. This acidic environment enables enzymatic DNA growth. At the same time, the outer electrode consumes diffusing protons, preventing the low-pH region from spreading to neighboring sites.

By repeating this process across selected locations during each synthesis cycle, the chip can build many different DNA sequences simultaneously.

The work builds on a semiconductor platform originally developed for neuroscience applications.

“A defining feature of the chip was precision current injection, which we used to permeabilize neuronal membranes for intracellular access,” said Donhee Ham, John A. and Elizabeth S. Armstrong Professor of Engineering and Applied Sciences at Harvard.

“At a certain point, we wondered whether that same current control could be redirected from cells to molecules – replacing the neuron-facing electrodes with ring-electrode pairs that could localize pH for DNA synthesis. It worked.”

Data storage potential

Beyond biological applications, the researchers explored whether the technology could support DNA-based data storage.

Using the 64 synthesized DNA sequences, the team encoded a 169-byte text message, demonstrating the platform’s potential for storing digital information in DNA molecules.

The researchers believe a water-based synthesis approach could become increasingly important if DNA production scales dramatically in the future.

“DNA data storage asks DNA synthesis to operate at a scale far beyond today’s needs,” said Woo-Bin Jung, co-first author of the study.

“That is why enzymatic synthesis in water can matter. If far more than 64 sequences can be synthesized in parallel, it could offer an environmentally friendly route toward writing DNA at very large scale.”

The team also attempted to increase the density of synthesis sites on the chip. While those experiments failed, they revealed that the main limitation came from the chemistry used to remove blocking groups rather than from the chip’s electronic architecture.

“The chip did what we asked it to do: it localized low pH at selected sites,” said Han Sae Jung, co-first author of the study. “The limitation came from the deprotection chemistry, not from the silicon.”

The study was published in Nature Electronics.

American startup PsiQuantum has begun construction on a facility in Moreton Bay, Australia, where it plans to build and deploy what it calls the world’s first utility-scale, fault-tolerant quantum computer.

The California-based quantum computing company broke ground on the site alongside representatives from the Australian government, the Queensland government, and local officials. The project is expected to become a key part of Australia’s push to establish itself as a hub for advanced computing technologies.

The facility will be developed in multiple phases and will house the infrastructure needed to support PsiQuantum’s photonic quantum computing architecture. One of the first major components arriving at the site will be a large cryogenic plant designed to cool the quantum system.

PsiQuantum said the cryoplant, being built by Linde Engineering, is expected to be delivered in the second half of 2027. The company ordered the system in late 2024 and described it as one of the largest cryogenic cooling plants ever built for a quantum computing application.

Race toward utility

The cooling system will be followed by cryogenic cabinets that will contain photonic quantum chips. These cabinets will be connected using conventional optical fiber, forming the foundation of the company’s planned utility-scale machine.

Unlike many quantum computing efforts that remain focused on laboratory-scale systems, PsiQuantum is pursuing a photonics-based approach designed to scale using existing semiconductor manufacturing processes and optical networking technologies.

“Building a quantum computer that solves real world problems is one of the great engineering challenges of our time. For decades, quantum computing has held the promise of transforming what humanity can achieve through computation, and today in Australia we are beginning to turn that promise into reality,” said Victor Peng, Chief Executive Officer of PsiQuantum.

The company believes fault-tolerant quantum computers could eventually tackle problems beyond the reach of conventional computing systems. Potential applications include drug discovery, materials development, energy research, logistics, manufacturing, agriculture, and financial modeling.

PsiQuantum also sees quantum computing as a complement to emerging computational technologies rather than a replacement. The company said future industrial computing systems are likely to combine quantum and classical technologies to accelerate scientific discovery and solve increasingly complex problems.

Building local capability

The Moreton Bay facility is also expected to support workforce development in Australia’s growing quantum technology sector. The site sits within the Moreton Bay Central Innovation Precinct, close to a TAFE Centre of Excellence and the University of the Sunshine Coast’s Moreton Bay campus.

According to the company, the project will create jobs in engineering, technical operations, advanced manufacturing, research, and professional services. The location is intended to help students, technicians, apprentices, and engineers move directly into careers linked to quantum technologies.

“Australia has been part of this journey from the very beginning,” said Prof. Jeremy O’Brien, Co-Founder and Executive Chair of PsiQuantum.

The groundbreaking follows the opening of PsiQuantum’s Test and Validation Lab at Griffith University in Brisbane in May 2026. The laboratory houses cryogenic equipment used to test photonic quantum chips and supporting systems that will eventually be deployed at the Moreton Bay site.

If completed as planned, the facility will become one of the world’s most ambitious efforts to build a fault-tolerant quantum computer at industrial scale.

The U.S. Department of Defense (DoD) has awarded Parallel Works, an Illinois-based software company spun out of Argonne National Laboratory, a contract to provide a unified platform that connects military supercomputing centers with secure commercial cloud infrastructure.

The contract, granted under the department’s High Performance Computing Modernization Program (HPCMP), will allow scientists, engineers, and acquisition professionals across the DoD to access both on-premises and cloud-based computing resources through a single interface. The goal is to speed up the development and deployment of high-performance computing (HPC) and artificial intelligence (AI) workloads used for defense research and operations.

Parallel Works said its ACTIVATE High Security Platform (HSP) will act as the control plane linking Defense Supercomputing Resource Centers (DSRCs) with commercial cloud services. The platform is designed to let users move workloads across environments while maintaining security requirements for sensitive data.

The company said researchers will be able to test and deploy workloads on emerging cloud infrastructure before those capabilities are integrated into the DoD’s supercomputing centers.

Connecting defense computing

The platform has been approved at Impact Level 5 (IL5), one of the highest security classifications for non-classified DoD cloud environments. According to the company, it is among a small number of software platforms approved to handle export-controlled workloads, including International Traffic in Arms Regulations (ITAR) and Controlled Unclassified Information (CUI).

The system is intended to address growing demand for computing power driven by AI development, simulation workloads, and digital modernization programs across the military.

“AI-driven warfare and the ramp to digital modernization are demanding far more model-sharing options than legacy infrastructure can provide,” said Keith Obenschain, Chief Technology Officer at HPCMP.

The platform offers on-demand access to cloud compute resources, allowing users to avoid traditional queue delays associated with shared supercomputing systems. It also enables organizations to expand computing capacity by distributing workloads across multiple environments and cloud providers.

Parallel Works said users will have access to cloud infrastructure from providers including Amazon Web Services, Microsoft Azure, Google Cloud, and Oracle Cloud.

Accelerating mission deployment

As part of the contract rollout, the Naval Research Laboratory has already implemented the platform to support weather forecasting workloads.

According to the company, the system automates forecasting workflows while securely coordinating computing resources across defense and cloud environments. The approach is intended to improve reliability, speed up processing, and help redistribute workloads when demand spikes.

“The HPCMP contract allows our platform to support a broad range of mission-critical HPC and AI workloads across the DOD teams,” said Matthew Shaxted, CEO of Parallel Works.

The company said the environment can also serve as a secure testing ground for AI development tools and next-generation cloud architectures before they are adopted within the DoD’s existing supercomputing infrastructure.

The contract reflects a broader push by the U.S. military to combine traditional supercomputing resources with commercial cloud services as AI models and data-intensive applications continue to increase computing requirements across defense operations.

KAIST researchers have developed a simulation-based method to predict how small future transistors can become before quantum effects begin to interfere with their operation, a breakthrough that could help engineers design next-generation semiconductor chips more efficiently.

The team used atom-level quantum mechanical calculations to identify the scaling limits of transistors, the tiny switches that control the flow of electricity in electronic devices. The findings could help chipmakers continue shrinking transistors beyond current technology nodes while reducing costly trial-and-error during development.

As the semiconductor industry enters the so-called 2 nm era, the physical dimensions of transistors remain significantly larger than 2 nanometers. One of the biggest obstacles to further miniaturization is quantum tunneling, a phenomenon in which electrons leak through barriers that would normally block them, making it difficult to control current flow.

Understanding where that limit lies has been challenging because it is nearly impossible to directly measure the atomic-scale interactions occurring where metal contacts connect to semiconductor channels.

Chasing atomic limits

To overcome that challenge, the KAIST team relied on first-principles calculations, a computational approach that predicts material behavior using the laws of physics rather than experimental data.

Building on a previously developed framework known as multi-space constrained-search density functional theory (MS-DFT), the researchers conducted virtual transfer length method experiments, a standard technique used to measure contact resistance between metal electrodes and semiconductor materials.

The simulations allowed the team to examine how electrons move across metal-semiconductor interfaces and determine the critical tunneling length, the point at which electron leakage begins to affect transistor performance.

The researchers applied the method to monolayer molybdenum disulfide (MoS2), a two-dimensional semiconductor considered a promising candidate for future transistor channels because it can be manufactured at atomic-layer thickness.

Their analysis showed that electron penetration into the channel varies depending on the choice of metal electrode and the atomic structure of the contact interface. As a result, the minimum achievable transistor size is not fixed but depends on material selection and device design.

Beyond today’s nodes

According to the study, the critical tunneling length changes based on the work function of the metal and the geometry of the contact structure. This means engineers can potentially tune transistor scaling limits by selecting different materials and interface configurations.

Among the combinations studied, the team found that electron leakage could be suppressed at dimensions below 4 nanometers, suggesting future transistors may be scaled even further than current technologies allow.

The researchers also proposed a design strategy that combines two-dimensional semiconductors with different properties to reduce power consumption in future chips.

“This study is significant because it presents a new physical criterion for defining how small next-generation transistors can become,” said Professor Yong-Hoon Kim.

“By computationally analyzing quantum mechanical phenomena in the sub-10 nm regime, which are difficult to probe experimentally, we have opened a path toward utilizing these findings in next-generation transistor design.”

The team believes the approach could provide chip designers with a platform for predicting transistor performance and scaling limits before fabrication begins, potentially shortening development cycles for future AI and high-performance computing chips.

The study was published in the journal npj Computational Materials.

Genesis AI has unveiled Eno, its first general-purpose robot, as the company pushes toward building machines that can work across factories, laboratories, hospitals, hotels and eventually homes.

The robot combines Genesis AI’s custom hardware with GENE, the company’s robotics foundation model, which serves as Eno’s control system. According to the company, the robot is designed to perform complex tasks, adapt to changing environments and complete objectives with minimal human intervention.

Genesis AI said Eno differs from traditional industrial robots because it is built to operate as a physical agent rather than execute a fixed set of commands. The company claims the system can understand goals, reason through changing conditions, retain memory and carry out multi-step tasks over extended periods.

The announcement comes as companies across the robotics sector race to develop humanoid and general-purpose robots capable of handling a wider range of real-world jobs with less specialized programming.

Built beyond humanoids

Instead of closely resembling a human, Eno uses a wheeled base topped by an adjustable tower structure. The robot can change its height and reach in real time and fold into a more compact form when not in use.

Its most human-like features are its robotic hands. Genesis AI says the hands match the form and function of human hands, allowing Eno to use existing tools and interact with environments already designed for people.

“The only path to creating a robot that can truly deliver value to society and excel in the real world is through intentional design and a single, comprehensive system,” said Zhou Xian, co-founder and CEO of Genesis AI.

“From day one we’ve approached our design and engineering through a production mindset built around bringing our hardware, software and intelligence together as a whole.”

The company said Eno was designed alongside GENE so that the robot’s body and intelligence system function as a single platform. GENE is intended to give the robot human-level dexterity and the ability to handle long-horizon tasks that require planning and adaptation.

Showing robotic intent

One of Eno’s distinguishing features is an optional built-in screen that displays the robot’s reasoning process, operational status and intended actions in real time.

Genesis AI said the feature is designed to improve transparency and help people understand what the robot is doing as it works alongside human operators.

“What Genesis is building with Eno is a fundamentally new model for extending human capability through advanced robotics,” said Eric Schmidt, former Google CEO and Genesis AI investor.

“The breakthrough is not replacing human expertise, but amplifying it — making advanced robotics genuinely useful, accessible, and scalable across industries.”

According to the company, Eno could eventually manage entire workflows rather than perform isolated actions. Examples include keeping production lines supplied, coordinating tasks across facilities and preparing workspaces for the next shift.

Genesis AI plans to begin production and targeted customer deployments before the end of 2026. Initial deployments will focus on manufacturing, logistics and laboratory environments, followed by hospitality and healthcare settings. Consumer-focused applications are expected at a later stage.

The company recently raised $105 million in seed funding and says additional robotic platforms are already under development as part of its broader roadmap for general-purpose robotics.

💾

SpaceX has agreed to acquire coding startup Cursor in a $60 billion all-stock deal, according to a filing with the U.S. Securities and Exchange Commission (SEC). The acquisition deepens the Elon Musk-led company’s push into artificial intelligence software and developer tools just days after its record-breaking Nasdaq debut.

The deal brings San Francisco-based Anysphere, the company behind Cursor, under the SpaceX umbrella as thje company looks to strengthen its position against rivals such as OpenAI and Anthropic in the fast-growing AI coding market.

SpaceX announced the massive deal in a post on X, saying: “SpaceX has exercised the option to acquire @cursor_ai in an all-stock transaction with the goal of building the world’s most useful AI models.”

SpaceX has exercised the option to acquire @cursor_ai in an all-stock transaction with the goal of building the world’s most useful AI models.

For the past few months, SpaceXAI has been jointly training a model with Cursor, which will be released in Cursor and Grok Build soon.… https://t.co/X5mepgXgjJ

— SpaceX (@SpaceX) June 16, 2026

The company added that it has been jointly training a model with Cursor over the past several months and plans to release it in both Cursor and Grok Build, xAI’s coding agent, in the near future.

Code meets compute

The acquisition comes after an agreement between the two companies in April that gave SpaceX the right to acquire Cursor at a predetermined valuation. According to SpaceX’s SEC filing, the companies also entered into a compute partnership under which SpaceX would provide GPU cluster capacity and collaborate on improving existing models, including Grok, while potentially developing new AI products together.

In the filing, SpaceX described software development as a strategically important AI use case because it generates structured, verifiable data and frequent user feedback. The company said Cursor’s coding workflows provide valuable developer interaction data, including coding prompts, iteration cycles and software architecture decisions that could help improve model training and inference.

Cursor has emerged as one of the fastest-growing AI software companies since its founding in 2022. Reports suggest that the company has reached roughly $2.6 billion in annualized business-to-business revenue and has recently been discussing a funding round that would have valued it at $50 billion.

The startup’s coding assistant is widely used by developers to generate, edit, and review software code. Cursor counts major companies such as Stripe, Adobe and Nvidia among its users, while Nvidia CEO Jensen Huang has described it as his “favourite enterprise AI service.”

Data powers expansion

SpaceX said the acquisition will help advance its broader AI strategy by integrating its models more directly into developer workflows and expanding distribution through high-engagement software products.

“We look forward to working closely with the Cursor team to advance our frontier AI capabilities,” SpaceX said in its announcement.

Cursor also welcomed the deal, posting on X: “We’re excited to join forces with @SpaceX to advance the frontier of useful AI. Expect significant improvements to Cursor soon.”

We're excited to join forces with @SpaceX to advance the frontier of useful AI. Expect significant improvements to Cursor soon. https://t.co/62IMr2sgEy

— Cursor (@cursor_ai) June 16, 2026

The acquisition follows SpaceX’s blockbuster stock market debut last week. Reuters reported that the company’s valuation climbed above $2 trillion after the listing. The company’s shares surged further Tuesday, briefly pushing SpaceX past Amazon by market value.

SpaceX expects the transaction to close during the third quarter of 2026, subject to regulatory approvals. Once completed, the deal would give the company direct control of one of the industry’s most prominent AI coding platforms while expanding access to developer data that it believes can strengthen future versions of Grok and other AI models.

Researchers in Australia have developed a neuromorphic vision chip that can see, process, and store visual information in a single device, mimicking how the human eye and brain work together. The small chip, built using doped indium oxide, is designed to reduce reliance on external processors and enable faster decision-making in applications such as autonomous systems.

The work was led by engineers at RMIT University, with contributions from Deakin University and the University of Melbourne. The team says the device combines sensing, processing, and memory functions in one platform, removing the need for separate hardware components that typically slow down machine vision systems.

Unlike conventional imaging systems that capture data and send it to external processors, the new chip performs computation directly where the light is detected. The sensing layer, which is thousands of times thinner than a human hair, is engineered to respond to light and retain information over time, allowing it to function more like a biological visual system.

The researchers say this integrated approach could reduce energy consumption and improve response speed in real-time environments. The device was tested using ultraviolet light, and the team is now working to extend its capabilities to visible and infrared light for broader applications.

Brain-like vision system

The chip is designed to mimic the way the human eye captures light and how the brain processes and stores visual input. It performs multiple tasks on a single platform, including sensing incoming light, processing signals, and storing visual information for later use.

Team leader Professor Sumeet Walia said the goal was to remove the delay and energy cost of transferring data between separate systems. “We’ve made real-time decision making a possibility with our invention, because it doesn’t need to process large amounts of irrelevant data and it’s not being slowed down by data transfer to separate processors.”

The device also showed the ability to retain visual information for longer periods without frequent electrical refresh signals, which reduces energy use and improves efficiency.

First author and RMIT PhD researcher Aishani Mazumder said the system draws inspiration from how the brain processes information. “Neuromorphic vision systems are designed to use similar analog processing to the human brain, which can greatly reduce the amount of energy needed to perform complex visual tasks compared with today’s technologies.”

Toward autonomous machines

The researchers say the technology could be used in self-driving cars, autonomous robots, and monitoring systems that operate in dangerous environments. Possible applications include object recognition in vehicles, detection systems in remote or hazardous areas, and advanced imaging for forensics and industrial inspection.

Because the chip integrates multiple functions into a single element, it could also support longer-term autonomous operation without heavy computational infrastructure. The team says this makes it suitable for systems that need to adapt quickly to changing environments.

The device mimics the retina’s ability to capture an entire image and the brain’s ability to interpret and store it, enabling a more compact and efficient approach to machine vision. The researchers believe this could eventually lead to vision systems that improve with experience, similar to biological systems.

The team used specialized nanofabrication and microscopy facilities at RMIT University, with support from the Australian Research Council and the National Computational Infrastructure.

The study appears in the journal Advanced Functional Materials.

Canadian-based critical minerals company E3 Lithium has commissioned Phase 2 of its Clearwater Project Demonstration Facility in Alberta and confirmed lithium concentrations of 75.8 milligrams per liter in brine produced during ongoing reservoir testing.

The company said the latest results support the consistency of lithium-bearing brine chemistry across the Clearwater Project and the broader Bashaw District, a key factor as it advances plans for commercial lithium production.

The demonstration facility is currently being used to test reservoir performance over an extended period. Brine is pumped from a production well and immediately reinjected into a separate injection well, allowing engineers to evaluate flow behavior and reservoir response under operating conditions.

According to the Alberta-based company, the facility achieved stable production and injection through the well pair. During early operations, an electrical submersible pump maintained the targeted flow rate of 1,400 cubic meters per day.

Brine tests confirm consistency

The company regularly collected brine samples during operations and sent them to an independent third-party laboratory for analysis. Results confirmed lithium concentrations of 75.8 mg/L, matching expectations for the resource.

E3 said the reservoir data generated through the testing campaign will be used to update its reservoir development plan and support a feasibility study for the Clearwater Project.

The next stage of operations will focus on analyzing the relationship between brine and natural gas within the reservoir. The information is expected to support the design of the front end of a future commercial processing plant.

The demonstration facility forms part of E3’s broader effort to reduce technical risks before moving toward large-scale lithium production.

“With Phase 2 operations underway, E3 has begun a series of project-scale de-risking activities for the Clearwater Project,” said Chris Doornbos, CEO and chair of E3 Lithium.

The company added that the current campaign has already processed more brine through the two new wells than all of its previous testing programs combined.

Scaling toward production

Lithium demand continues to grow as automakers and battery manufacturers seek additional supplies of the metal used in electric vehicle batteries and energy storage systems.

For E3, the demonstration facility serves as a bridge between laboratory-scale testing and commercial deployment. Data collected from reservoir operations, brine chemistry studies, and future process testing will be used to refine engineering designs and operating plans.

“We have already moved more brine through these two wells than E3 has handled through all its previous testing combined, and we expect to continue to operate the pump as needed over the coming 24 months,” Doornbos said.

He added that confirmation of reservoir performance and lithium grades at the new wells strengthens confidence in the predictability of the resource.

The company plans to continue operating the demonstration facility while advancing front-end engineering design work and completing a feasibility study for the project. The resulting data will help determine the design of a future commercial facility and support decisions related to full-scale development.

China has launched a new research laboratory dedicated to photonic computing, a technology that uses light instead of electricity to process information and could help address the growing power demands of artificial intelligence and advanced computing.

The Shanghai Key Laboratory of Integrated Photonic Computing Chips and Systems was officially launched this week in Shanghai. According to local reports, it is China’s first industry-academia platform focused specifically on photonic computing research and development.

The facility is based at Shanghai Jiao Tong University and will bring together researchers and industry partners to develop photonic chips, optical components, computing architectures, and software systems needed to support the technology.

The launch comes as countries and technology companies race to secure enough computing power to train and run increasingly sophisticated AI models. As data centers consume more electricity, researchers are looking beyond traditional semiconductor designs for alternatives that can deliver higher performance with lower energy use.

Light over electrons

Unlike conventional chips that move data using electrons through silicon circuits, photonic chips rely on photons, or particles of light, to transmit and process information.

Because photons travel faster than electrons and generate significantly less heat, photonic systems could potentially offer major gains in speed, bandwidth, and energy efficiency. Researchers believe such systems may help overcome some of the physical limitations facing traditional semiconductor technologies.

Zou Weiwen, director of the new laboratory and a professor at Shanghai Jiao Tong University, highlighted the technology’s potential, describing it as “an important pathway for achieving breakthroughs in computing power, offering advantages in bandwidth, latency, and energy efficiency.”

The new laboratory will focus on several areas, including photonic chip architectures, silicon-photonics integration, optical devices, supporting algorithms, and commercial applications.

China has increasingly identified photonic computing as a strategic technology as it seeks alternatives to advanced semiconductor technologies that remain difficult to access because of export restrictions and supply chain challenges.

Building local ecosystem

Shanghai authorities said they have supported research into photonic computing through multiple science and technology programs while also investing in innovation platforms and industrial development.

The push toward photonic computing comes as researchers worldwide explore alternatives to conventional chip designs that are approaching physical and thermal limits. Supporters of the technology argue that light-based processors could help reduce the massive energy consumption associated with AI data centers while enabling faster data movement across computing systems.

The laboratory is a joint initiative between Shanghai Jiao Tong University and Lightelligence, a Shanghai-based startup specializing in photonic computing technologies.

Lightelligence has previously claimed to be the first company to achieve large-scale deployment of hybrid optical-electronic computing systems. The company has emerged as one of China’s leading players in the photonic computing sector.

Despite growing interest in the technology, significant technical hurdles remain before photonic computing can be deployed at scale.

Zou noted that one major challenge is the lack of a mature software and algorithm ecosystem capable of fully utilizing photonic hardware. Researchers must also address fundamental scientific and engineering problems before the technology can achieve its theoretical performance advantages.

Still, the launch of the new laboratory signals China’s intention to accelerate research into light-based computing as demand for high-performance computing and AI infrastructure continues to grow worldwide.

Researchers at Tohoku University have developed an iron-based catalyst that improves the performance of zinc-air batteries, a technology viewed as a lower-cost alternative to lithium-ion systems for energy storage.

The team focused on overcoming a key limitation of the oxygen reduction reaction (ORR) in zinc-air batteries. While zinc-air batteries draw oxygen from the surrounding air instead of storing it internally, the ORR process is relatively slow, restricting overall battery performance.

To address the problem, the researchers engineered a heterointerface combining iron oxide (Fe2O3) and samarium oxide (Sm2O3). The interface altered how electrons behave at the catalyst surface, helping speed up reaction kinetics and improve battery efficiency.

According to the researchers, the catalyst weakens excessive bonding between iron sites and hydroxyl intermediates, allowing the reaction to proceed more smoothly and reducing one of the major bottlenecks in zinc-air battery operation.

Reworking battery chemistry

Fe2O3 was selected because it is abundant, inexpensive and stable under the alkaline conditions commonly found in zinc-air batteries. However, iron-based catalysts typically suffer from strong Fe-OH interactions that slow the release of reaction products.

The newly developed Fe2O3/Sm2O3 interface was designed to overcome that limitation. Researchers said charge redistribution, orbital hybridization and spin modulation at the interface helped optimize the catalyst’s behavior during oxygen reduction.

As a result, the catalyst delivered faster ORR kinetics and improved durability compared with conventional approaches. The work also avoided the use of expensive noble metals, which are commonly used in high-performance catalysts.

The team evaluated the material in both liquid and flexible all-solid-state zinc-air batteries. Beyond laboratory measurements, the researchers demonstrated practical operation by powering a small LED lamp and charging a smartphone.

“The catalyst achieves high ORR activity, improved reaction kinetics, excellent durability, and superior performance in both liquid and flexible all-solid-state zinc-air batteries,” said Hao Li, distinguished professor at the Advanced Institute for Materials Research (WPI-AIMR).

Beyond precious metals

Zinc-air batteries have attracted attention because they use widely available materials and can theoretically deliver high energy density. Unlike lithium-ion batteries, they rely on oxygen from the atmosphere as part of the electrochemical process.

Researchers believe improving catalyst performance could help make the technology more practical for real-world applications, including portable electronics, wearable devices, and large-scale energy storage systems.

“For the general public, this means that we may be closer to affordable and sustainable clean-energy technologies,” Li said.

The researchers say the study demonstrates a potential pathway toward lower-cost energy storage by replacing precious-metal catalysts with more abundant materials while maintaining strong electrochemical performance.

Future work will focus on expanding the interfacial spin-regulation strategy used in the study and exploring whether similar approaches can improve other energy-conversion and storage technologies.

The findings were published in Angewandte Chemie International Edition.

The Department of War has released a third batch of declassified records tied to unidentified anomalous phenomena (UAP), adding 72 files that describe unexplained sightings reported by military personnel, federal agents, and foreign observers.

The release is part of the PURSUE program, launched after President Donald Trump directed federal agencies to identify, review, and declassify records related to UFOs, UAPs, and potential extraterrestrial activity.

The newly released documents do not provide evidence of alien life or a government cover-up. Instead, they offer additional details about several unresolved cases that investigators have been unable to conclusively explain.

According to the Department of War, unresolved cases are incidents where available data is insufficient to determine the nature of the observed object or phenomenon.

Strange objects, few answers

One of the most detailed reports involves a February 2022 sighting near Colorado Springs, Colorado. Five US Army personnel at Fort Carson reported seeing an unusual object hovering above Cheyenne Mountain.

“The object was ‘potato’ shaped with distinct edges and appeared to look painted in a creamy/whitish opalescent color,” an FBI document states. Witnesses said the object appeared to consist of “articulating fish scales or panels that were non-symmetrical, non-overlapping and irregular shaped.”

According to the report, the object remained stationary for roughly two minutes before disappearing. No photographs or videos were captured.

Investigators later assessed with “low confidence” that the sighting may have resulted from sunlight reflecting off snow-covered terrain and illuminating clouds. However, witnesses reported clear skies, and investigators found no evidence of balloons, aircraft, or known foreign technology in the area.

Another case involved six federal law enforcement agents who reported repeatedly observing bright orange and red orbs in October 2023. According to the documents, the objects appeared above a ridgeline and, in some instances, seemed to generate additional smaller orbs.

An analysis conducted by the Pentagon’s All-Domain Anomaly Resolution Office reviewed several possible explanations, including military exercises and flare deployments occurring nearby. The report said those explanations were plausible but not conclusive and noted that the observations could involve “unrecognized technology.”

Global reports surface

The latest release also includes an FBI report describing a 2025 sighting in the northeastern United States. A witness reported seeing a bright red sphere near a wooded area.

“Inside the red sphere, at the center, there appeared to be what (redacted) described as a ‘white plasma sun’ about the size of a basketball,” the report states.

The witness said a second sphere appeared before both objects departed silently. The file includes cellphone footage showing two glowing red objects moving across the sky. The White House later shared the video online while referencing the case.

Among the older documents is a CIA report from Zimbabwe dating to 2008. Witnesses near the country’s main airport described a disc-shaped object with rotating lights.

“At one point during observation, ‘beams’ were observed emanating from the object,” the report states.

So far, the Trump administration’s disclosure effort has resulted in the release of roughly 300 UAP-related files dating back to the 1940s. The Department of War said additional records will be released in future tranches as reviews and declassification efforts continue.

Northrop Grumman has developed a new gallium nitride (GaN) microwave chip that operates in the W-band spectrum, a high-frequency range used for advanced radar, satellite communications, and emerging wireless networks.

The company said the chip was designed, manufactured, and made market-ready in less than six months, an unusually short timeline for a high-frequency semiconductor intended for defense and commercial applications.

Built at Northrop Grumman’s semiconductor facility in Redondo Beach, California, the chip is designed to strengthen wireless signals while operating at frequencies capable of transmitting large amounts of data with low latency.

According to the company, the technology could support secure military communications, next-generation radar systems, satellite links, and future 5G and 6G wireless networks.

Faster signals, smaller hardware

The W-band occupies frequencies between roughly 75 and 110 gigahertz, offering significantly more bandwidth than lower-frequency wireless systems. The added spectrum can help enable faster data transmission, higher-resolution sensing, and more secure communications.

W-band technologies are attracting growing interest because they can handle far larger amounts of data than many conventional wireless systems. The frequencies are increasingly being explored for applications ranging from high-resolution imaging and advanced sensing to satellite communications and future mobile networks.

However, operating at such high frequencies requires specialized semiconductor materials and designs capable of maintaining performance while managing power and heat, areas where GaN-based devices have gained increasing attention.

Northrop Grumman said the new chip fully utilizes this portion of the radio spectrum to improve signal performance while reducing the need for larger and more power-intensive equipment.

Gallium nitride has become a key material for advanced radio-frequency electronics because it can operate at higher power levels and frequencies than conventional silicon-based devices. The material is already widely used in military radars, electronic warfare systems, and satellite communications hardware.

The company said the chip can act as a signal amplifier, helping wireless systems transmit and receive data more efficiently. In military applications, that could improve secure communications and data transfer between platforms and satellites. Commercially, the technology could help support future wireless networks as demand for higher data rates continues to grow.

Defense To 6G

The project was developed through the California DREAMS hub of the Microelectronics Commons program, a public-private initiative aimed at accelerating domestic semiconductor innovation.

The program is funded by the Office of the Under Secretary of Defense for Research and Engineering and brings together industry, government, and academic partners to speed the development of advanced microelectronics technologies.

Northrop Grumman said the rapid development cycle demonstrates how collaborative manufacturing and design efforts can shorten the path from concept to deployment.

The company manufactures microelectronics at two US-based facilities and produces millions of semiconductor components annually for defense and commercial customers. Its Microelectronics Center has been involved in semiconductor development for more than seven decades.

As governments and industries seek faster communications, more capable radar systems, and advanced sensing technologies, high-frequency GaN devices are expected to play an increasingly important role in future wireless infrastructure.

Qcells has begun manufacturing solar cells at its new facility in Cartersville, Georgia, bringing the company closer to operating what it says is the United States’ first and only fully vertically integrated solar manufacturing factory.

The company announced that the plant is now producing solar cells and expects all production lines to reach full capacity by the third quarter of 2026. Once fully operational, the facility will manufacture ingots, wafers, cells, and solar modules under one roof.

The start of cell production marks a significant milestone for domestic solar manufacturing, as most solar panels installed in the US still rely on imported components. Qcells said the Cartersville site will become the largest operating solar cell factory in US history.

Module assembly operations at the facility are already running at full capacity, producing about 16,700 solar panels every day.

Building domestic supply chains

When fully ramped up, the Cartersville factory will produce 3.3 gigawatts (GW) of annual capacity each for ingots, wafers, and cells, along with 3.5 GW of solar modules.

Combined with Qcells’ expanded manufacturing facility in Dalton, Georgia, the company’s total US module production capacity is expected to reach 8.6 GW annually by the end of the third quarter. According to the company, that output is enough to generate the equivalent annual electricity needs of roughly 1.3 million American homes.

“Producing the first solar cells at Cartersville is a milestone for Qcells and for American manufacturing,” said Andy Park, Global CEO of Qcells.

The company said producing key solar components domestically could provide customers with greater certainty around supply availability, pricing, and potential tariff-related disruptions.

Tax credits and jobs

Qcells said the Cartersville operation could also help project developers qualify for the federal Domestic Content Bonus tied to the Investment Tax Credit. Because major solar module components are produced in the US, developers may find it easier to meet domestic sourcing requirements.

The company added that its vertically integrated manufacturing approach allows it to claim production incentives under Section 45X of the Advanced Manufacturing Production Tax Credit across multiple stages of the solar supply chain, including ingots, wafers, cells, and modules.

The investment is also expected to support thousands of manufacturing jobs in Georgia. Together, the Cartersville and Dalton facilities are projected to employ nearly 4,000 workers, including about 3,800 direct jobs across Bartow and Whitfield counties.

Qcells said the Cartersville factory is the first US facility of its kind built in more than a decade and includes what it describes as the largest ingot and wafer production plant ever constructed in the country.

Researchers have developed a soft robotic gripper that can inspect fruit for ripeness and harvest it without causing damage, a capability that could help reduce food waste and improve harvesting efficiency.

The device uses soft silicone and polyurethane fingers equipped with multiple sensors to measure a fruit’s size, shape, color and firmness before deciding whether it is ready to be picked. The system was tested on strawberries and can remove the fruit by twisting its stem rather than cutting it.

The gripper was developed by a team that includes Anand Mishra, assistant professor in the Department of Mechanical, Materials and Aerospace Engineering at West Virginia University. The research began at Cornell University and is continuing at WVU.

The technology addresses a major challenge in agriculture: harvesting delicate fruits at the right stage of ripeness. Fruits such as strawberries and raspberries have a narrow harvest window and can bruise easily during picking, transportation and storage.

Gentle grip, smart sensing

“Our gripper’s quick, accurate inspections and harvesting can reduce spoilage of fruits and lower supply chain costs,” Mishra said.

“Fruit inspections are critical for harvesting decisions that have traditionally been made by human workers. However, using human workers for harvesting involves challenges such as labor shortages, health concerns and inaccuracies in picking.”

Agricultural robots have been proposed as a solution, but many existing systems are designed for controlled greenhouse environments rather than outdoor farms. Traditional robotic grippers can also damage soft fruit because of their rigid structures.

The new system combines tactile and visual sensing in a soft robotic design. Stretchable optical fibers embedded inside each finger act as tactile and curvature sensors, while a miniature camera and distance sensor are mounted in the palm.

The five-fingered gripper can identify a fruit’s shape, stiffness and ripeness while also detecting slippage during handling. Researchers reported that the device can open and close in less than two seconds, lift loads weighing up to one kilogram and achieve nearly 100 percent accuracy in shape prediction.

According to the team, the technology could be particularly useful for fruits that do not show obvious visual signs of ripeness. Avocados are one example, where growers often rely on touch rather than appearance to determine harvest timing.

Beyond the orchard

“My research group at WVU is called the Robiotics Lab because we focus on robots that mimic biology,” Mishra said.

“We design squishy, squeezy, rubbery robots inspired by animals.”

The gripper’s design draws inspiration from biological systems and resembles both a human hand and a starfish. Researchers say the soft structure allows it to absorb forces differently than conventional rigid robots, making it better suited for handling delicate objects.

Beyond agriculture, the sensing technology could find use in other fields that require careful manipulation.

“This system offers applications in space exploration, health care, food handling and underwater manipulation,” Mishra said.

“In biomedical robotics, for instance, the integration of curvature and tactile sensing could enhance wearable and rehabilitation devices.”

Researchers believe the combination of soft materials and integrated sensing could help robots interact more safely and effectively with both objects and their surroundings.

The study was published in the journal Nature Communications.

US President Donald Trump said Thursday that he has canceled planned military strikes against Iran, claiming that negotiations between Washington and Tehran have reached a stage where a broader agreement is close to being finalized.

The announcement marked a dramatic shift after days of escalating tensions that saw both countries exchange strikes and threats, raising fears of a wider conflict in the Middle East.

In a post on Truth Social, Trump said discussions had advanced significantly and received backing from multiple countries involved in the negotiations.

“Based on the fact that discussions with the Islamic Republic of Iran have been brought to the highest level of Iranian leadership and approved, I have, as President of the United States of America, cancelled the scheduled strikes and bombings against Iran this evening,” Trump wrote.

He also claimed that the final elements of the agreement had been approved by several regional and international stakeholders, including Israel, Saudi Arabia, the UAE, Qatar, Turkey, Pakistan, Bahrain, Kuwait, Jordan and Egypt.

From threats to talks

The decision came just hours after Trump issued fresh warnings against Tehran, threatening to hit Iran “very hard tonight” and suggesting the United States could take “total control” of Iran’s oil and gas markets, including the strategic Kharg Island export hub.

Despite calling off the strikes, Trump said US pressure on Iran would continue until a formal agreement is signed.

“The Naval Blockade will remain in full force and effect until this Transaction is finalized – Time and place of the signing to be announced shortly,” he wrote.

The Strait of Hormuz remains one of the world’s most important energy chokepoints, with a significant portion of global oil shipments passing through the waterway. Any disruption there has the potential to impact international energy markets and shipping routes.

Trump later told reporters that a signing ceremony could take place in Europe as early as this weekend. According to reports, Vice President J.D. Vance is expected to attend on behalf of the United States.

Hormuz remains under watch

“The strait (of Hormuz) will officially open as soon as we sign, which could be soon — very soon, maybe over the weekend in Europe,” Trump said. “I won’t be able to be there, but JD will be there.”

Iran has not officially confirmed Trump’s claims that a peace deal is imminent.

However, reports indicate that Qatari officials recently visited Tehran to discuss possible pathways toward a broader agreement aimed at reducing tensions between the two countries.

Trump has for months signaled that a breakthrough with Iran was near. The latest claims come after a week of tit-for-tat strikes that strained an already fragile ceasefire reached in April.

Meanwhile, Iran’s parliamentary speaker Mohammad Bagher Ghalibaf warned that further escalation could have long-term consequences for global energy markets.

He said that “wrong strategies and impulsive decisions” could destabilize energy supplies and “create an endless quagmire that you will be stuck in for years.”

Stellantis has integrated Factorial Energy’s solid-state battery cells into a Dodge Charger Daytona development vehicle and begun road testing the technology, marking a key step toward bringing solid-state batteries into production vehicles.

The milestone is the first time Factorial’s FEST (Factorial Electrolyte System Technology) solid-state cells have been installed in a Stellantis vehicle. The companies will now evaluate battery performance, reliability, and safety under real-world driving and charging conditions.

The development follows laboratory validation completed in 2025, when Stellantis and Factorial demonstrated automotive-scale FEST cells with an energy density of 375 Wh/kg. The cells also achieved charging from 15 percent to 90 percent in 18 minutes and operated across temperatures ranging from -22 °F to 113 °F.

Moving from laboratory testing to an on-road vehicle required engineers to redesign parts of the battery system. Stellantis said it developed a patented mechanical architecture that allows the solid-state cells to fit within an existing battery pack while maximizing performance.

From lab to road

Engineers also modified battery controls and pack design to meet automotive requirements for durability and safety. According to the companies, these changes were needed to ensure consistent performance under a range of driving and charging conditions.

The battery pack is based on Stellantis’ STLA Large platform and serves as the first vehicle in a previously announced multi-stage development program.

“Battery development is a balancing act. It’s not enough to optimize a single metric. We need a system that delivers real benefits in a real vehicle,” said Ned Curic, Stellantis Chief Engineering and Technology Officer.

“This milestone shows we are bringing solid-state batteries closer to our customers with the potential for longer range, faster charging, and lower costs. Just as important, FEST’s strong compatibility with lithium-ion manufacturing processes gives us a critical path to scale this technology.”

Solid-state batteries replace the liquid electrolyte used in conventional lithium-ion batteries with a solid material. The technology is expected to improve energy density, increase safety, and reduce charging times. However, scaling production and integrating the cells into vehicles have remained major challenges for automakers and battery developers.

Scaling battery technology

With the battery now installed in a working vehicle, Stellantis and Factorial have launched a testing and calibration program to gather performance data from real-world operation.

The companies said the program will evaluate charging behavior, vehicle safety, and long-term reliability. The results will also help refine battery management systems and pack-level controls before any future commercial deployment.

“We are deeply honored to work alongside Stellantis, one of the world’s great mass-market automakers, on this STLA Large-based development car,” said Siyu Huang, CEO of Factorial.

“What we have built together, from cell chemistry to pack architecture to enable real-world road testing, is exactly the kind of deep, full-stack collaboration that solid-state has always required.”

Factorial, which is backed by Stellantis, Mercedes-Benz, Hyundai and Kia, is among several companies racing to commercialize solid-state batteries. The latest milestone moves its technology beyond cell-level testing and into vehicle validation, a critical stage before potential mass-market adoption.

A Chinese startup has claimed a breakthrough in photonic chip manufacturing by producing 8-inch optical chip wafers without relying on conventional deep ultraviolet (DUV) lithography systems.

Hangzhou-based Prinano said it successfully validated mass production of photonic chips using its nanoimprint lithography (NIL) technology, a manufacturing approach that physically stamps nanoscale patterns onto wafers instead of projecting them through optical systems.

According to the company, the process was carried out in collaboration with Shenzhen Litra Technology and used its PL-AS vacuum air-cushion nanoimprint lithography equipment. Prinano said the method “completely avoids” the need for DUV lithography tools while reducing manufacturing costs to about one-tenth of traditional solutions.

The announcement comes as Chinese chipmakers continue looking for alternative manufacturing technologies amid restrictions on access to advanced lithography systems from Dutch supplier ASML.

Seeking lithography alternatives

Photonic chips are used in applications such as optical communications, sensing systems, lidar, and data centers. Unlike conventional electronic chips, they process information using light rather than electrical signals.

Prinano said its technology supports wafer-level production of photonic chips and can be applied to optical communication and sensing devices based on gallium arsenide, indium phosphide, and silicon photonics.

The company said its PL-AS tool offers sub-10-nanometer line-width resolution along with wafer-level pressure control. However, it did not disclose production volumes, manufacturing yields, or customer shipment data.

Nanoimprint lithography has long been viewed as a potential alternative to traditional optical lithography because it can create extremely fine features without the need for expensive projection optics. The technology was pioneered in the 1990s by Princeton University professor Stephen Y. Chou.

Prinano founder Ge Haixiong studied under Chou and established the company in 2017 to develop NIL equipment, materials, and manufacturing processes.

Questions over scalability

Interest in nanoimprint lithography has increased in recent years as chipmakers seek lower-cost manufacturing methods. In 2023, Canon introduced a commercial NIL system and promoted it as an alternative to some advanced lithography tools.

Despite the promise, analysts have questioned whether the technology can match the throughput and yields required for large-scale semiconductor manufacturing.

Research firm SemiAnalysis has previously noted that while NIL equipment can be cheaper than advanced lithography tools, overall economics depend on factors such as defect rates, template production, throughput, and process integration.

Prinano’s announcement focuses specifically on photonic chips rather than advanced logic processors, a distinction that may make commercialization more achievable. Photonic devices often rely on repeating nanoscale structures such as gratings, waveguides, and ring resonators, which can be well suited to imprint-based manufacturing.

The company said the technology has moved into wafer-level production validation, but independent verification of yields, commercial deployment, and manufacturing scale has not yet been disclosed.

For China’s semiconductor industry, the development highlights ongoing efforts to explore alternative manufacturing routes as access to advanced lithography equipment remains restricted.

Google has unveiled DiffusionGemma, a new experimental AI model that generates text using diffusion rather than the autoregressive approach used by most large language models today. The company says the model can deliver up to four times faster text generation on dedicated GPUs while running on consumer hardware.

The model builds on Google’s Gemma 4 family and Gemini Diffusion research. Unlike traditional language models that generate text one token at a time from left to right, DiffusionGemma creates and refines blocks of text in parallel.

According to Google, the approach enables output speeds exceeding 1,000 tokens per second on an NVIDIA H100 GPU and more than 700 tokens per second on an NVIDIA GeForce RTX 5090.

The company says DiffusionGemma is aimed at developers working on speed-sensitive applications such as interactive editing, rapid content iteration, code infilling, and other workflows where low latency is more important than maximum output quality.

Breaking text bottlenecks

Most large language models generate text sequentially, predicting one token after another. While effective, this process can leave local hardware underutilized when serving a single user.

DiffusionGemma takes a different approach. Instead of generating text word by word, it creates a 256-token block at once and then repeatedly refines it through multiple passes.

Google compares the difference to moving from a typewriter to a printing press. Rather than waiting for each token to be generated before producing the next one, the model processes an entire section of text simultaneously.

The company says this shifts the bottleneck from memory bandwidth to compute performance, allowing modern GPUs to operate more efficiently during local inference.

Another key feature is bi-directional attention. Since the model generates text in parallel, every token can attend to every other token during generation. This makes it better suited for tasks where future context matters, such as code completion, in-line editing, mathematical structures, and biological sequences.

Google highlighted a demonstration in which DiffusionGemma was fine-tuned to solve Sudoku puzzles, a task that can be challenging for conventional autoregressive models because later tokens influence earlier decisions.

Designed for local AI

The model uses a 26-billion-parameter mixture-of-experts architecture but activates only 3.8 billion parameters during inference. According to Google, this allows the model to fit within roughly 18 GB of VRAM when quantized, making it accessible on high-end consumer GPUs.

DiffusionGemma also includes an iterative self-correction mechanism. Because it evaluates an entire text block during refinement, it can identify and fix mistakes as generation progresses.

However, Google acknowledged that the model prioritizes speed over quality. The company said standard Gemma 4 models remain the preferred choice for production environments where output quality is the primary concern.

The speed advantage is also most apparent in local deployments and low-concurrency environments. In cloud settings serving large numbers of users simultaneously, conventional autoregressive models can often utilize hardware efficiently through batching, reducing the benefits of diffusion-based generation.

Google has released DiffusionGemma under an Apache 2.0 license through Hugging Face and is supporting deployment through tools including MLX, vLLM, Hugging Face Transformers, NVIDIA NeMo, and Unsloth.