Skip to content

Invasive Electrodes

Invasive (intracortical) electrodes are inserted directly into the cortex and can simultaneously record spikes + LFP from hundreds to thousands of neurons. They remain the highest-quality BCI acquisition method to date, and they are the main technical route for Pitt, BrainGate, and Neuralink.

1. Three Generations of Electrode Technology

Generation Representative Channels Features
1st: rigid silicon Utah Array 96–128 4×4 mm², 1–1.5 mm shanks
2nd: high-density silicon probes Neuropixels 1.0 / 2.0 384 / 5120 Long thin shanks (70×20 μm), thousands of sites per shank
3rd: flexible / high-throughput Neuralink N1, Paradromics CSA 1024–4096 Polymer flexible threads, reduced tissue response

2. Utah Array

Utah Array (Blackrock Microsystems) is by far the most widely used intracortical electrode in clinical practice:

  • Specs: 10×10 shank grid (96 channels), 400 μm inter-shank pitch, 1–1.5 mm shank length, 4×4 mm² footprint
  • Materials: Silicon substrate + platinum/iridium electrode tips
  • Implantation: Pneumatic inserter — a single pulse drives the array into cortex
  • History: First human implant in 1997; used by the BrainGate program from 2004 to the present

Representative studies: - Hochberg et al. 2006 (BrainGate-1 first implant in a paralyzed patient) - Collinger et al. 2013 (Pitt 7-DoF robotic arm) - Willett et al. 2021/2023 (handwriting & speech BCI)

3. Neuropixels

Neuropixels, developed jointly by the Allen Institute + HHMI + UCL + IMEC in 2017, is a high-density silicon probe:

  • Specs: Single shank 10 mm × 70 μm × 24 μm, with 960 recording sites along the axis and 384-channel live switching
  • Advantages: A single shank records hundreds of units, covering multiple brain regions in depth
  • Disadvantages: Rigid shank, not suited to long-term human implantation (currently used mainly in animal research)

Neuropixels 2.0 (2021): - 4-shank configuration, 5120 total sites - Widely used for IBL (International Brain Lab) cross-lab standardized recording

Neuralink N1 (2024–2026 PRIME study) represents the new flexible-electrode direction:

  • Specs: 1024 channels × 64 flexible threads, 16 electrodes per thread
  • Implantation: "Sewn" by the R1 robot (< 100 μm thread width avoids blood vessels)
  • Advantages:
    • Flexible threads reduce tissue response (relative to Utah's rigid shanks)
    • Robot-precision placement avoids blood vessels
    • Wireless BLE transmission
  • Known issues (publicized 2024-05): Post-op thread retraction in patient Noland Arbaugh left ~15% of electrodes usable; the team compensated algorithmically to restore >8 bps ITR

5. Paradromics Connexus

Paradromics (FDA Breakthrough Device Designation, 2025-11) with its Connexus Direct Data Interface: - Specs: 4 "cortical modules" (~420 electrodes each) + wireless transcutaneous data link - Total channels: ~1680 - Positioning: Sells throughput rather than single-electrode resolution as the key value - Indication: ALS-related loss of communication

6. Floating Arrays

To reduce tissue-electrode relative motion (brain pulsation from respiration + cellular response), a class of floating arrays decouples the electrode base from the skull, letting it "float" on the cortex:

  • Cortec Vector Array: silicone substrate + polymer electrodes
  • Argo / SiNAPS: active CMOS integration
  • Flatiron (Starlab): preclinical flexible floating

7. Core Electrode Performance Metrics

Metric Definition Typical (Utah) Frontier (Neuralink)
Channel count Number of independently recorded sites 96 1024+
Yield Fraction of channels yielding a unit 60–80% 30–50% (early)
SNR Spike amplitude / noise RMS 5–20 10–30
Long-term stability Usable channels at 6 months ~40% Insufficient data
Tissue response Glial sheath thickness around electrode 50–100 μm < 20 μm (target)

8. Tissue Response and Long-Term Stability

Foreign body response is the central bottleneck for long-term stability of intracortical electrodes:

  1. Acute phase (0–7 days): Blood-brain barrier disruption, microglial activation
  2. Chronic phase (7 days – months): Astrocytes encapsulate the electrode, forming a glial sheath
  3. Electrode failure: The sheath isolates the electrode from neurons, degrading SNR

Directions forward: - Shrink electrode size: Flexible < 100 μm threads (Neuralink) - Coatings: PEDOT conductive polymer, slow-release anti-inflammatory coatings - Materials: Carbon fiber, graphene

9. Coupling with Decoding Algorithms

Channel count × decoder capability jointly determine BCI performance:

  • Utah 96 channels + Kalman: 2D cursor control, 6-DoF robotic arm (BrainGate 2004–2012)
  • Utah 192 channels + RNN: 62-WPM speech BCI (Willett 2023)
  • Neuralink 1024 channels + undisclosed decoder: >8 bps cursor control (PRIME 2024)

Hardware progress drives decoder upgrades: high-throughput electrodes + Transformer foundation models is the primary research direction post-2025.

10. Logical Chain

  1. Utah Array is the clinical standard, but 96 channels and tissue response cap its ceiling.
  2. Neuropixels transformed animal research, making "hundreds of units per shank" standard.
  3. Neuralink / Paradromics push flexible high-throughput, aiming for 1000+ long-term stable channels.
  4. Long-term stability is the core engineering problem — coatings, materials, and size shrinkage proceed in parallel.
  5. Channel count × decoder capability = BCI performance; neither alone is sufficient.

References

  • Normann et al. (1999). A neural interface for a cortical vision prosthesis. Vision Res. — Origin of the Utah Array
  • Jun et al. (2017). Fully integrated silicon probes for high-density recording of neural activity. Nature. — Neuropixels https://www.nature.com/articles/nature24636
  • Musk & Neuralink (2019). An integrated brain-machine interface platform with thousands of channels. J Med Internet Res. https://www.jmir.org/2019/10/e16194/
  • Paradromics (2025). Connexus Direct Data Interface FDA Breakthrough Designation. Press release.
  • Polikov et al. (2005). Response of brain tissue to chronically implanted neural electrodes. J Neurosci Methods. — Review of tissue response

评论 #