Bosonic error correction is a family of techniques that encodes a logical qubit into the states of a harmonic oscillator, such as a microwave cavity mode, rather than into a fixed two-level system. Because the oscillator has an infinite-dimensional Hilbert space, error syndromes like photon loss can be detected and corrected without immediately collapsing the encoded information. The patent set here covers both the correction protocols themselves and the measurement techniques needed to read out these bosonic modes with high fidelity.

The assignee ranking covers 11 companies, and it is heavily concentrated: the top five combined account for 86.9% of the 61 records in scope, and the top ten reach 98.4%. The single largest filer holds 24 records, well ahead of the rest of the ranked group. This is a field where a small number of organisations, spanning both established technology companies and university-affiliated filers, hold nearly the entire documented patent position.

Filings rose from zero in 2017 to a peak of 26 records in 2022, then declined using the last two complete years on record: 2021 had 6 filings and 2024 had 4, a -33% change over that span. Because publication typically lags filing by around 18 months, 2025 and later are still incomplete and should not be read as a continued decline. The honest read is that the initial wave of foundational filings has passed its peak, not that interest in the technology itself is fading.

G06N, covering computing arrangements based on AI models, appears on 88.5% of the 61 records in scope, making it by far the dominant classification. Hardware-adjacent classes such as G06F, B82Y, H10N, H01L, H01S, H10D and G01N each appear on a much smaller share, from roughly 5% to 15% of records. Because a single record can carry several IPC classes, these shares add up to more than 100%, and the pattern shows that most claim activity is being drafted as computational method rather than as physical device or material innovation.

The thinnest IPC branches relative to the dominant computing cluster are G01N (material analysis and testing) at 4.9% of records, and H01S (lasers and stimulated emission) and H10D (semiconductor devices, general) each at 6.6%. These branches sit well below the 88.5% share held by G06N, meaning diagnostic, characterisation, and specific device-implementation claims are far less crowded than the core correction-algorithm space where the leading filers already hold dense claim positions. A new filer is better placed pairing a narrow hardware or measurement claim with these thinner classes than contesting the crowded computational-method core directly.