The M3 (Michigan Micro Mote) computer alone, shown floating above what appears to be Lincoln's shoulder for scale. (Photo taken at the Michigan Integrated Circuits Laboratory. Photographer: Michael Simari)
The future is now…and it’s tiny.
In A Nutshell
- Researchers built autonomous robots just 210-340 micrometers wide, roughly the size of a paramecium, that contain an onboard computer, sensors, memory, and propulsion systems.
- The microrobots can measure temperature, make decisions based on sensor data, and navigate toward warmer areas without any external control, all while running on the same power as a living cell.
- About 100 robots fit on a chip smaller than a fingertip and are manufactured using standard semiconductor processes, with estimated production costs of about one penny per robot.
- The devices demonstrated the ability to autonomously adapt their movement in response to changing temperature gradients, transmitting measurement data back to researchers by encoding it in their motion patterns.
Robots the size of a single-celled organism can now sense their environment, make decisions, and act on them without any outside help. Researchers at the University of Pennsylvania and University of Michigan created microscopic machines measuring just 210 to 340 micrometers wide (roughly the size of a paramecium or two human hairs laid side by side) that pack in an onboard computer, temperature sensors, memory, communication systems, and propulsion.
Published in Science Robotics, the work marks the first reported demonstration of a fully integrated, task-specific onboard computer, environmental sensors, and locomotion systems in something this microscopically small. The robots operate without external control. These devices run on roughly 100 nanowatts of power, about the same energy budget as many living cells.
Previous attempts at building robots at cellular scales forced researchers to sacrifice key capabilities. Most microrobots either relied on external equipment like magnetic coils to control them, could only execute predetermined behaviors hard-coded during manufacturing, or lacked the ability to sense and respond to their surroundings. These new microrobots overcome all three limitations.
Complete robot next to the year on a penny for scale. (Credit: Kyle Skelil, University of Pennsylvania)
Built Like Computer Chips, Small as Cells
The research team manufactured their microrobots using the same semiconductor processes that create computer chips. About 100 robots fit on a single millimeter-scale chip that can rest on a gloved fingertip. Each individual robot contains a tiny processor, solar cells for harvesting power from light, temperature sensors, circuits for controlling movement, memory, and an optical receiver for wireless programming.
Power represents the primary constraint when working at cellular dimensions. Living cells have evolved molecular machinery to harvest and use energy efficiently at nanowatt levels. The research team had to match this biological efficiency. The processor alone consumes nearly 90% of the robot’s 100-nanowatt power budget and occupies about 25% of its body.
To work within these cellular-scale power limits, researchers designed a custom computer architecture that compresses robot actions into specialized instructions. Commands like “sense the environment” or “move for N cycles” execute in what appears to be a single operation. This compression makes meaningful tasks possible with just a few hundred bits of memory.
A microrobot, fully integrated with sensors and a computer, small enough to balance on the ridge of a fingerprint. (Credit: Marc Miskin, University of Pennsylvania)
The robots demonstrated their autonomous capabilities through experiments that mirror how single-celled organisms navigate their environments. In one test, microrobots continuously measured surrounding temperature, converted readings to digital data, and transmitted results back to a base station by encoding information in their movement patterns.
When tested in a bath of gradually warming solution, the robots’ measurements matched those from standard temperature probes. The sensors achieved 0.3-degree Celsius resolution with about 0.2-degree accuracy despite their microscopic size. This performance exceeds most existing digital thermometers of comparable volume.
The second experiment tested whether robots could exhibit taxis, or directed movement toward or away from environmental stimuli that characterizes many microorganisms. Researchers programmed the microrobots to search for warmer regions when temperature dropped, then hold position when finding warmth.
Results showed responsive behavior driven by real-time sensor input. Robots initially rotated in place without an imposed gradient. When researchers cooled the local area, robots automatically switched to exploratory movements, traveling through their environment until locating warmer zones, where they resumed stationary rotation. Reversing the temperature gradient caused robots to reverse course. This showed responses to live environmental changes rather than a fixed movement script.
Moving at cellular scales requires different approaches than standard robotics. The robots use electrokinetic propulsion, passing current between oppositely charged platinum electrodes while immersed in fluid. Mobile ions surrounding the robot respond to this electric field. The ions drag fluid along, creating flow that propels the machine forward at 3 to 5 micrometers per second. Robots can travel in different directions or rotate by changing which electrodes are active.
The microbots are produced in a sheet (top left) roughly the area of a fingertip (bottom left). Each bot contains solar cells for harvesting energy, some of which also double as optical receivers, a temperature sensor on each side of the microbot for detecting differences, a processor for taking in information and making decisions, four actuator panels that drive its motion. Four of the receivers allow the robot to identify whether an incoming program is addressed to it. (Credit: Maya Lassiter, University of Pennsylvania)
Getting instructions into robots the size of cells required wireless solutions. The team developed an optical communication system using light-emitting diodes to both power and program the devices. One LED wavelength provides energy that solar cells convert to electricity. A second wavelength transmits data by flashing patterns that robots interpret as binary instructions and write to onboard memory.
A graphical user interface automates the entire programming process. Researchers can define robot behaviors without writing low-level firmware code. The system can send initialization programs to configure basic functions or task programs that define operations. Once instructions load, the robots operate completely autonomously based on their internal program and sensor readings.
To prevent random light fluctuations from accidentally altering robot behavior, the communication protocol requires passcode sequences. Each robot recognizes both a global passcode common to all devices and a type-specific code for addressing particular subsets. This enables researchers to give different instructions to different robots, similar to how cells in a multicellular organism respond to different chemical signals.
Microbots released after fabrication using microelectronics approaches. The cost per robot can be under a dollar, and they may be programmed individually or as a group. (Credit: Maya Lassiter, University of Pennsylvania)
Potential Medical Applications and Future Improvements
The robots’ ability to sense, process, and respond to temperature could support future applications in biological research and medical diagnostics. Operating at cellular scales, they probe thermal gradients in ways larger sensors cannot, fitting into microfluidic chambers or capillary tubes where traditional instruments fail.
The devices could potentially interface with living systems by positioning their aqueous environment near target tissues and allowing heat to flow between environments. Reading temperature without direct physical contact bypasses biocompatibility concerns that affect many implantable sensors. The current work demonstrates these capabilities in controlled laboratory conditions, not in living organisms.
The authors estimate that, at production scale, each robot could cost on the order of a penny. Combined with the simple programming system requiring only controllable light sources rather than specialized laboratory equipment, this low cost could make cellular-scale autonomous robotics accessible beyond well-funded research institutions.
The research team notes that more advanced applications will require improvements like new actuators for in-body operation or better power transfer methods. Moving to more advanced semiconductor processes would increase onboard memory about 100-fold, enabling programs approaching thousands of lines of code and supporting more sophisticated decision-making.
For decades, roboticists have worked to shrink autonomous machines while maintaining the key features that distinguish robots from simpler devices: onboard sensing, programmable computation, and independent decision-making. These microrobots achieve all three at dimensions where they can coexist with the cellular building blocks of life itself.
VIDEO: Sequence shows microscopic robot performing a sequence of drive, turn, and slide states.
The current robots face several constraints inherent to their microscopic scale. Memory is limited to a few hundred bits due to leakage currents in the 55-nanometer CMOS process used for fabrication. Propulsion speed remains relatively slow at 3-5 micrometers per second because operating voltage sits below the optimal range for electrokinetic propulsion. The robots require operation in fluid environments, specifically 5 millimolar hydrogen peroxide solution in the experiments conducted. Temperature sensing is the only fully demonstrated sensor modality, though the design includes an electric field sensor not extensively characterized in the published work. The optical communication and power system requires controlled illumination between 200 and 2,600 watts per square meter. Operation inside living organisms has not been demonstrated.
This work received support from the National Science Foundation (NSF 2221576), the University of Pennsylvania Office of the President, the Air Force Office of Scientific Research (AFOSR FA9550-21-1-0313), the Army Research Office (ARO YIP W911NF-17-S-0002), the Packard Foundation, the Sloan Foundation, and the NSF National Nanotechnology Coordinated Infrastructure Program (NNCI-2025608), which supports the Singh Center for Nanotechnology. The authors declared no competing interests.
Authors: Maya M. Lassiter (Department of Electrical and Systems Engineering, University of Pennsylvania), Jungho Lee (Department of Electrical Engineering and Computer Science, University of Michigan), Kyle Skelil (Department of Mechanical Engineering and Applied Mechanics, University of Pennsylvania), Li Xu (Department of Electrical Engineering and Computer Science, University of Michigan), Lucas Hanson (Department of Physics and Astronomy, University of Pennsylvania), William H. Reinhardt (Department of Electrical and Systems Engineering, University of Pennsylvania), Dennis Sylvester (Department of Electrical Engineering and Computer Science, University of Michigan), Mark Yim (Department of Mechanical Engineering and Applied Mechanics, University of Pennsylvania), David Blaauw (Department of Electrical Engineering and Computer Science, University of Michigan), and Marc Z. Miskin (Department of Electrical and Systems Engineering, University of Pennsylvania, corresponding author). Published in Science Robotics, Volume 10, article number eadu8009, December 10, 2025. DOI: 10.1126/scirobotics.adu8009. Submitted February 10, 2025; Accepted November 13, 2025.