Inside Paradromics’ Connexus BCI and the Speech Restoration Frontier

When we discuss the current state of implantable Brain-Computer Interfaces (BCIs), the public conversation is easily dominated by flashier, software-heavy headlines. There is a common assumption that if we just throw more advanced neural networks and larger language models at the brain, we can decode complex human thought from almost any signal.

But anyone working in neuroengineering knows that a decoding algorithm is only as good as its underlying data. If you feed a state-of-the-art AI a low-resolution, noisy stream of brain activity, you will get slow outputs.

Austin-based Paradromics is tackling this with a hardware-first approach. Their flagship commercial device, the Connexus® BCI, is built on a specific architectural wager: the ultimate limit of a neural interface isn't the AI decoding the data, it is the raw bandwidth of the physical hardware.

The Core Philosophy: Resolving Individual Action Potentials

To understand why Connexus is designed the way it is, we have to look at how different BCIs gather data. Some systems sit safely on top of the scalp (non-invasive EEG), others sit directly on the surface of the brain (ECoG). To decode fluent, natural human speech, you need to isolate individual voices, specifically, the single-neuron action potentials (spikes) in the motor cortex responsible for speech movements.

Paradromics maximizes this resolution by optimizing for Information Transfer Rate (ITR). In their recent pre-clinical benchmarks using the SONIC framework (Standard for Optimizing Neural Interface Capacity), the Connexus system achieved a massive data throughput of over 200 bits per second (bps) at the latency of 56 milliseconds. To put that in perspective, that is over 20 times faster than older, traditional intracortical research arrays.

Inside the Hardware Architecture

The Connexus BCI is a fully implantable, Class III medical device designed to remain invisible and function safely inside a human body for years. The system breaks down into three interconnected components:

[Cortical Module](Brain) ---> (Flexible Extension Lead) ---> [Internal Transceiver (Chest)]                                                                                                          

1. The Cortical Module

This module is placed directly by a neurosurgeon onto the target area of the motor cortex.Protruding from its titanium-alloy body are 421 platinum-iridium microwires.

• The Dimensions: Each wire is micro-fabricated to be less than 40 microns in diameter, thinner than a strand of human hair.

• The Material Choice: Paradromics chose platinum-iridium and titanium. These are the exact materials used for decades in deep brain stimulation (DBS) systems and pacemakers because they resist corrosion and won't cause adverse tissue reactions.

• The Channel Scalability: While a single module houses 421 channels, the broader Paradromics platform is architected to link up to four modules together, opening the door to over 1,600 recording channels in future applications.

2. The Internal Transceiver

The tiny signals captured by the microelectrodes pass through patented, on-chip processing directly inside the module. From there, a thin, highly flexible extension lead routes the data out of the skull, under the skin, down to a disk-shaped internal transceiver implanted in the patient's chest.

3. Power and Telemetry

The chest transceiver does not use an onboard chemical battery that requires surgical replacement. Instead, it relies on an inductive charging link, similar to how a modern smartphone charges wirelessly on a pad. An external transceiver worn by the user powers the internal device through the skin while establishing a secure, high-speed optical data link to send the neural features to an external computer for real-time AI decoding.

The brain is soft, gelatinous, and highly protective. When a rigid object is anchored to the skull while the brain naturally floats and shifts beneath it, microscopic shearing forces occur. This friction triggers an immune response known as glial scarring, where the brain builds a protective wall of scar tissue around the electrodes, insulating them and killing the neural signal within months or years.

The Connexus cortical module solves this by floating naturally with the brain. Because the entire module is tiny and sits directly on the cortex without being bolted to the skull, it moves in tandem with the tissue. Combined with the very small cross-sectional area of the 40-micron microwires, this approach minimizes tissue displacement and mechanical trauma. Pre-clinical studies have already showed over three years of rock-solid neural signal stability.

The Connect-One Study

The FDA recently granted Paradromics an Investigational Device Exemption (IDE) to launch their first-in-human clinical trial: the Connect-One Study.

This study is an Early Feasibility Study (EFS) designed specifically for adults (ages 22 to 75) who have severe motor limitations and loss of speech due to conditions like Amyotrophic Lateral Sclerosis (ALS), brainstem strokes, or spinal cord injuries.

Patient Selection & Screening (Prerequisite Phase)

Candidates with severe dysarthria (slurred speech) or anarthria (total speech muscle loss) who retain cognitive clarity are screened across three primary US clinical sites (including UC Davis and the University of Michigan).

Neurosurgical Implantation (Standard Surgical Workflow)

A neurosurgeon implants the cortical module, subcutaneous leads, and chest transceiver. Crucially, the system uses established neurosurgical workflows, avoiding the need for a highly proprietary surgical robot.

System Calibration & Training (Months 1–15)

Participants work closely with the Connect-One research team several times a week, right from their homes. Advanced language models and custom machine learning pipelines are trained to map the 421 channels of motor intent directly into synthesized speech, text, or cursor control.

Long-Term Longitudinal Tracking (Years 2–6)

After the primary 15-month evaluation phase, the study tracks the participant for an additional 5 years to evaluate the ultimate goal of the system: proving that a high-bandwidth BCI can maintain long-term safety and high performance for a user's lifetime.

Paradromics has taken a pragmatic, deeply academic route. By relying on highly mature biocompatible materials, a high-density microwire footprint, and an architecture that prioritizes raw data throughput, the Connexus BCI feels less like science fiction and more like a highly refined, inevitable evolution of modern medical hardware.

References:

He, D., Siok, W. T., & Wang, N. (2026). Toward robust, reproducible, and widely accessible intracranial language brain-computer interfaces: A comprehensive review of neural mechanisms, hardware, algorithms, evaluation, clinical pathways, and future directions. arXiv preprint arXiv:2603.12279. 

Jain, A. (2026). Brain-computer interface: An update for the clinicians. Frontiers in Human Neuroscience, 20, Article 1777024. https://doi.org/10.3389/fnhum.2026.1777024

Patrick-Krueger, K. M., Burkhart, I., & Contreras-Vidal, J. L. (2024). The state of clinical trials of implantable brain–computer interfaces. Nature Reviews Bioengineering, 3(1), 50–67. https://doi.org/10.1038/s44222-024-00239-5 

Perkins, S. M., Reitman, M. E., Jarosiewicz, B., Patel, A. N., Nishimura, K., Qiao, S., & Angle, M. R. (2025). SONIC: A benchmarking paradigm for brain-computer interfaces. bioRxiv. https://doi.org/10.1101/2025.09.30.679683