Rocket One Inc. has signed a one-year sponsored research agreement with Virginia Commonwealth University (VCU) to fabricate and test magnetic computing devices designed to perform AI-related operations with potentially lower energy consumption. The research will focus on straintronic magnetic tunnel junctions, nanoscale components that use voltage-induced mechanical strain to control magnetic states. Planned demonstrations include an ultra-low-power analog multiplier, a magnetic crosspoint device for matrix computation, and a compact XOR logic gate. The program targets early-stage hardware research rather than a production-ready AI accelerator, with experimental testing intended to evaluate whether the devices can translate theoretical efficiency estimates into measurable results.
Rocket One and VCU Explore Magnetic Hardware for Energy-Efficient AI
The race to expand artificial intelligence capabilities is increasing demand for computing hardware that can deliver more processing performance within constrained power and cooling budgets. While much of the AI infrastructure market relies on conventional semiconductor processors and accelerators, researchers are investigating alternative computing architectures that could reduce the energy required for specific operations.
Rocket One Inc. is pursuing one such approach through a sponsored research agreement with Virginia Commonwealth University. The one-year program, running from October 19, 2026, through October 19, 2027, will focus on fabricating and experimentally testing straintronic magnetic tunnel junctions for potential use in future AI chips and accelerators.
The project is led by VCU Professor Supriyo Bandyopadhyay, with planned fabrication collaboration involving Professor Jian-Ping Wang of the University of Minnesota. Its central question is whether nanoscale magnetic devices can perform selected computing functions with low energy consumption while retaining information without continuous power.
The announcement describes a research and device-demonstration effort. It does not establish that the proposed components are ready for integration into commercial AI processors.
How Straintronic Magnetic Computing Works
Straintronic magnetic tunnel junctions use mechanical strain induced by voltage to influence magnetic behavior. This approach differs from conventional electronic logic, which generally relies on controlling current through semiconductor transistors.
Magnetic tunnel junctions typically contain magnetic layers separated by a thin insulating barrier. Their electrical resistance depends on the relative orientation of the magnetic layers. In strain-mediated designs, voltage-induced strain can help control the magnetic state, potentially reducing the energy needed to switch or manipulate it.
One potential advantage is nonvolatility: a magnetic state can persist after power is removed. That property could support computing architectures in which information remains stored locally instead of requiring continuous energy to maintain a state.
However, the practical benefits depend on device characteristics, including switching reliability, operating speed, fabrication consistency, electrical behavior and the overhead associated with controlling and reading the devices. A low-energy operation at the individual-device level does not automatically translate into a more efficient complete processor.
Rocket One and VCU plan to evaluate three device concepts that address different parts of the computing process.
Three Demonstrations Target AI and Digital Logic
The first objective is an analog magnetic multiplier that accepts two voltage inputs and multiplies them. Multiplication is a foundational operation in machine learning, particularly in the matrix computations used by neural networks. If the device can perform this function with sufficiently low energy and adequate accuracy, it could inform the development of specialized AI hardware.
The second objective is a magnetic crosspoint device with continuously tunable, linear electrical conductance. Crossbar-based computing architectures use arrays of interconnected devices to perform operations associated with matrix-vector multiplication. Such architectures are being explored as a way to move computation closer to where data is stored.
For these systems, predictable conductance and linearity can matter because device variation and nonlinear behavior may affect computational accuracy. The research will also investigate ways to expand the proposed device’s operating range. The announcement does not specify the accuracy, speed or array size that the program is expected to achieve.
The third demonstration targets a compact exclusive-OR (XOR) logic gate based on a single straintronic magnetic tunnel junction. XOR operations are used in arithmetic circuits, error detection and cryptographic functions. A single-device implementation could be useful if it can meet practical requirements for reliable switching, integration and operation.
Together, the demonstrations explore analog computation, crossbar-compatible devices and digital logic. They are distinct technical goals, and success in one would not by itself establish the commercial viability of the others.
The 100-Attojoule Estimate Needs Experimental Validation
The research plan cites a theoretical estimate of approximately 100 attojoules per multiplication operation for the proposed magnetic multiplier. An attojoule is one quintillionth of a joule, making the figure an indication of the very small energy scale targeted by the design.
That estimate should not be confused with a measured result from the newly announced program. The agreement calls for device fabrication and testing, and the company has not announced experimental energy measurements from this effort.
Comparisons with conventional AI accelerators would also require a consistent measurement basis. Device-level switching or multiplication energy is not directly equivalent to the total energy consumed by a processor, which includes data movement, control circuitry, memory, interconnects and other system components.
The practical test will be whether the devices demonstrate the predicted behavior under experimental conditions and whether their performance can be preserved when incorporated into larger circuits.
From Laboratory Devices to AI Infrastructure
Rocket One identifies AI infrastructure, autonomous systems, and defense and space applications as potential areas of interest. These environments can place a premium on compact processing and lower power consumption, particularly where local computing is necessary or power availability is limited.
Yet the pathway from a laboratory device to a deployable system involves substantial additional engineering. Researchers would need to demonstrate repeatable fabrication, characterize performance across operating conditions, integrate devices into functional circuits and evaluate reliability. Applications in space or defense would also require qualification for their intended environments.
The agreement gives Rocket One an option to negotiate licenses for intellectual property arising from the project, including the possibility of seeking an exclusive license within its commercial field, subject to the agreement’s terms and a separately negotiated license. The option provides a potential route toward commercialization but does not guarantee that licenses will be granted or that the technology will reach market.
For now, the project represents an early-stage attempt to translate magnetic computing research into measurable device results. Its significance will depend on what fabrication and testing establish about energy use, computational accuracy and the feasibility of integrating these components into future AI hardware.
Market Landscape
AI hardware development is increasingly focused on improving energy efficiency alongside raw computing performance. Conventional GPUs and other accelerators remain central to modern AI workloads, while alternative approaches—including in-memory computing, analog processing and emerging nonvolatile devices—seek to reduce the cost of moving and processing data.
Straintronic magnetic computing is a research-stage approach within this broader effort. Its potential advantages include nonvolatile storage and low-energy operation for selected functions. The key hurdles are demonstrating reliable devices, preserving computational accuracy, achieving scalable fabrication and showing that device-level efficiency survives at circuit and system level.
Rocket One’s agreement with VCU establishes a defined experimental program and a possible intellectual-property licensing pathway. It is too early to assess commercial competitiveness without measured results and system-level comparisons.
Top Insights
- Three device targets: The program will test an analog multiplier, a magnetic crosspoint device and a single-junction XOR gate.
- Theoretical efficiency remains unverified: The cited 100-attojoule estimate per multiplication operation is a theoretical figure, not a measured result from the new program.
- Nonvolatile operation is a potential advantage: Magnetic states that persist without continuous power could support future in-memory computing designs.
- Commercialization is a longer-term objective: Processor integration, manufacturing scalability and application-specific qualification remain additional steps.
- Licensing is not guaranteed: Rocket One has an option to negotiate intellectual-property licenses under the agreement’s terms, rather than an automatic exclusive license.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
