Firmware Release Notes
Firmware release notes for the Binho Supernova and the Binho Pulsar, newest first. They share one firmware line, so one set of notes covers both. The Atira and the Nova are listed under Other products. To install a version, see Updating Firmware.
| Version | Date | Summary | Use with |
|---|---|---|---|
| 4.7.1 | 2026-11-18 | I3C controller fixes, CDC UART fix, vendor CCC writes up to 1024 bytes | CosmicSDK / PyCosmicSDK 1.10.1 |
| 4.7.0 | 2026-11-10 | I3C target IBI and Hot-Join requests, SMBus writes as a target, external supply check on target voltage, I3C fixes | CosmicSDK / PyCosmicSDK 1.10.0 |
| 4.6.2 | 2026-10-23 | I2C clock and recovery, SPI and I3C fixes | CosmicSDK / PyCosmicSDK 1.7.1 |
| 4.6.1 | 2026-09-30 | HDR-DDR fix | CosmicSDK / PyCosmicSDK 1.6.0 |
| 4.6.0 | 2026-09-29 | I2C duty cycle, SMBus clock stretching, SPI clock change | CosmicSDK / PyCosmicSDK 1.6.0 |
| 4.5.1 | 2026-09-24 | SPI and I2C fixes | CosmicSDK / PyCosmicSDK 1.5.1 |
| 4.5.0 | 2026-09-16 | Transfers up to 32 KB, I2C and SPI target mode | CosmicSDK / PyCosmicSDK 1.5.1 |
Released 2026-11-18. Works with CosmicSDK 1.10.1 and PyCosmicSDK 1.10.1. CosmicSDK and PyCosmicSDK 1.10.0 also work; only the vendor CCC write needs 1.10.1.
Highlights
Section titled “Highlights”- Vendor CCC WRITE up to 1024 bytes. New command
I3C_CONTROLLER_VENDOR_CCC_WRITE(0x190F) for vendor CCCs (0x61to0x7Fbroadcast,0xE0to0xFEdirect), SDR only.I3C_CONTROLLER_CCC_TRANSFER(0x1909) is unchanged and stops at 255 bytes. Firmware 4.7.0 answersUNSUPPORTED_COMMAND. Measured: 256, 512 and 1024 bytes, 10/10 each, byte-equal on the wire.
Behavior Changes
Section titled “Behavior Changes”- A target whose IBI the device table NACKs, and that repeats the request right after each NACK, gets DISEC(INT) on the third NACK in a row with the requests less than 1 ms apart. Its dynamic address stays. Send ENEC to that target once its IBIs are accepted again. Targets that withdraw after one NACK keep the plain STOP.
- While a target holds SDA low, NACKed-IBI notifications stop at the controller’s retry limit (about 13). Firmware 4.7.0 sent about 200 per second. Real IBIs are still reported.
- An SDR read over 255 bytes below 7.5 MHz that the controller cannot end on time stops after about 10 to 25 ms and returns
SUCCESSwith the bytes received, ending with STOP. hold_scl_lowno longer clocks SCL before the hold. With clock stretching off, firmware 4.7.0 sent 18 calibration clocks right before the reset pulse.
- ENTDAA assigns every target again. Firmware 4.7.0 gave a dynamic address only to the first target. 10/10.
- SDR reads over 255 bytes end on time at every rate. At 7.5 MHz and above they timed out after 1 s, and a read ending on the target’s last byte left the bus without STOP. 1400/1400 over 7 rates and 5 lengths.
- A target in the device table that re-raises a NACKed IBI no longer resets the adapter. USB drop 10/10 to 0/10.
- CDC UART passthrough: a host writing faster than the UART no longer resets the adapter or drops bytes. USB now holds the host while the UART drains. Supernova at 115200 baud: 3/3 resets to 0/3.
- An ENTDAA right after a NACKed CCC no longer reports that NACK. 10/10.
Known Limitations
Section titled “Known Limitations”enable_i2c_clock_stretchingandenable_smbus_writesstill send 18 calibration clocks (about 3 µs period, SDA high, no START). Call them once at setup, not around each reset. With the mode on,hold_scl_lowsends no extra clocks.- The configured push-pull and open-drain rates are not restored after INIT_BUS, ENTDAA or Hot-Join DAA until the next I3C transfer.
- Pulsar CDC UART passthrough stops forwarding after about 1 KB when the target streams at 921600 baud.
Released 2026-11-10. Works with CosmicSDK 1.10.0 and PyCosmicSDK 1.10.0.
Highlights
Section titled “Highlights”- I3C target IBI and Hot-Join requests (Supernova). As an I3C target, the Supernova raises an IBI (MDB plus up to 7 payload bytes) or a Hot-Join when the host asks. SDR only. A request the bus state does not allow returns an error before anything goes on the bus, a code from
0x061Cto0x0620; a request the controller never answers ends with0x0621. Firmware 4.6.2 answersUNSUPPORTED_COMMAND. - SMBus writes as an I3C target, for MCTP over SMBus (Supernova). While on, a legacy I2C write sent as a target goes out with clock stretching and the target answers again right after the STOP, so the endpoint’s MCTP response is not NACKed during a role switch. Measured byte-exact at 100 kHz, 400 kHz and 1 MHz (runs at 746 kHz), 20/20 each.
- Raw I3C target layout
MEMORY_RAW_1024_BYTES(Supernova). No register address is taken from the bus, so an SMBus Block Write arrives whole. Writes are stored back to back. CosmicSDK and PyCosmicSDK 1.10.0 return an error for this layout on older firmware. - Zero-length I3C transfers (SMBus quick command). A 0-byte SDR or legacy I2C write sends START, address, STOP; a 0-byte read clocks one byte and discards it; a 0-byte HDR-DDR read returns
INVALID_PARAMETER. Before, a 0-byte write timed out after 1 s and a 0-byte read left the I3C engine busy until a reset. - On-adapter line capture for I3C image sensors. New command
I3C_CONTROLLER_CAPTURE_LINES(0x190E): the adapter reads each sensor line itself, on a GPIO edge or a timed line period, and sends the frame over USB bulk. Measured with a VD55G1 at 10 MHz: 160x120 frames 400/400. - Optional automatic read on a Pending Read IBI. Off by default. CosmicSDK and PyCosmicSDK support follows in a later release.
Behavior changes
Section titled “Behavior changes”- I3C
BUS_TIMEOUTrecovery now sends STOP and the target reset pattern, and keeps the host’s push-pull rate and drive strength. Before, a target left mid-message took the next transfer as its continuation, and STANDARD drive strength became FAST after the first timeout. - IBI payloads longer than 8 bytes are delivered as their first 8 bytes, with no truncation flag. Before, 8 bytes left the bus dead and 9 or more gave no notification.
- An IBI NACKed from an unknown address is ended with a CCC in the same frame: a broadcast RSTDAA when no I3C target is in the table (all dynamic addresses are cleared), otherwise a direct DISEC(INT) to that address. After an adapter reset, run INIT_BUS before talking to targets.
- A clamped target READ/WRITE event returns
0x0623(I3C_TARGET_TRANSFER_LENGTH_CLAMPED) instead of a 0-byte success. - I3C target memory layouts: data written before a repeated-START read now raises a WRITE event. Target READ events no longer report
ABORTED_BY_CONTROLLER; a read the controller ends with the T-bit isSUCCESS. - I2C clock stretching and SMBus writes as a target return
0x0622(I3C_I2C_SCL_RISE_TOO_SLOW) when SCL has no pull-up. Fit an external 1–4.7 kΩ pull-up to the I3C rail. - An unknown
TRIGGER_PATTERNvalue returnsINVALID_PARAMETER. Only 0 (target reset) and 1 (HDR exit) are accepted. FW_I3C_ENGINE_BUSY(0x0619) under IBI load is retryable. PyCosmicSDK 1.10.0 adds an optional time-bounded re-send (engine_busy_timeout).
- Target voltage after a reset. A saved 1.8 V target voltage no longer starts at about 3.3 V after a reboot, firmware update or DFU. Measured: 20/20 boots at 1814–1815 mV (Supernova) and 1785 mV (Pulsar).
- I2C pull-ups follow the rail. With the rail off, the adapter’s I2C pull-ups no longer back-power an unpowered target (81–85 mV → 2–4 mV at 2.2 kΩ).
- Level shifters in Nova-compat and terminal mode follow the rail, so a target with its rail off is not driven through them.
- I3C external supply check no longer fails every I3C Set because of the adapter’s own pins feeding an unpowered target (0/15 → 15/15 accepted).
- I3C controller init after a reboot answers while a self-powered target holds an IBI request from its old address; the adapter no longer resets (0/10 → 10/10).
- An I3C target that holds SDA low no longer resets the adapter. While SDA is held, I3C commands answer
I3C_ENGINE_BUSY(0x0619), and IBIs are served again once SDA is released. - I3C bulk transfers. One that returns an error for the current role no longer makes every later bulk transfer time out until a reset (this bug is in 4.6.2). One that hits the 1 s timeout answers
FW_BUS_TIMEOUTand frees the bulk pool. - I3C CCC WRITE carries up to 255 data bytes. Before, a write over 64 bytes could reset the adapter or corrupt the target table.
- I3C target read-from. Correct data after a read ended without STOP (with 7E up to 12.5 MHz; without 7E up to 10 MHz); no lost byte 8 or 9 and no bad first byte or NACK under IBI load; no NACK of its own address when a USB interrupt lands between the sub-address and the read header; a write after a read-from with repeated START reaches its own register.
- I3C target writes. The first write after target init is no longer lost. In the raw layout, a write aborted without STOP no longer merges with the next frame. Under bulk writes, command replies are no longer lost.
- IBI and Hot-Join. An IBI or Hot-Join on an idle bus after I3C init or RSTACT is served. NACKed IBIs under load no longer drop command completions. A Hot-Join NACKed because the table is full is not given an address and reports
HJ_REQUEST_REJECTED(0x0615). IBI payloads are no longer split across two notifications after a private read. - I3C controller reads. SDR reads longer than 8 bytes no longer stall. 1024-byte reads at 100 kHz no longer duplicate a byte in about 8 % of reads (see Known limitations).
- SDR, CCC or I2C after an HDR-DDR write with
SKIP_EXIT_PATTERNis answered; the controller sends the HDR Exit Pattern first. - Stalled DAC write. No longer leaves every system command,
GET_DEVICE_INFOincluded, timing out until a reset. It answersVOLTAGE_ADJUSTMENT_FAILUREafter 3 attempts.
Known limitations
Section titled “Known limitations”- I3C HDR-DDR in target mode is not supported.
- An external I3C HV supply above 3.3 V is accepted. The HV sense saturates at about 3299 mV.
- The 10-bit I2C target ACKs any second address byte.
- SPI controller reads above the wiring’s clock ceiling come back shifted by one bit while reporting success. See 4.6.1.
- In CAN-FD mode (Pulsar), large (bulk) transfers are not available.
- A target that re-requests a NACKed IBI right away keeps the I3C controller busy.
- After a read ended without STOP, a read-from to a Supernova I3C target at 12.5 MHz push-pull without 7E is NACKed about 1 time in 3. It never returns wrong data. Use 7E or 10 MHz or less.
- While a target holds SDA low, the host receives about 200 IBI-rejected notifications per second. The adapter keeps answering and recovers when SDA is released. A host loop that reads notifications until they stop should use a timeout.
- A write ended without STOP followed by a read-from to a Supernova I3C target is reliable up to 6.25 MHz push-pull; above that the read can return stale leading bytes.
- A Supernova I3C target that sees the target reset pattern a second time stores the next private write 2 bytes off. 4.6.2 does the same.
- A 1024-byte I3C read can still duplicate one byte while reporting
SUCCESS: about 0.1–0.5 % at 100 kHz, and about 0.6 % at 5 MHz with a 30 % duty cycle. At 5 MHz use the 50 % duty cycle, keep a read under about 1 ms of bus time, or check the data (PEC or CRC).
Released 2026-10-23. Works with CosmicSDK 1.7.1 and PyCosmicSDK 1.7.1. The external supply check on target voltage listed in earlier copies of these notes ships in 4.7.0.
Behavior changes
Section titled “Behavior changes”- I2C no longer resets the adapter on a stuck bus. When a target holds the bus, an I2C transfer or scan now ends with a timeout (about 1.5 s) instead of the adapter restarting about 8 s later.
- I2C bus recovery after a failed scan. A scan that fails at bus level now resets the controller and clocks up to 9 pulses on SCL to free a target holding SDA. The next transfer works without a reset.
- I2C STOP timing after a NACK. With the adapter’s pull-ups disabled, the STOP after a NACK now meets the I2C setup time (measured 5.3 µs; it was under 0.1 µs).
- 10-bit I2C scan. Scanning with 10-bit addresses no longer leaves the adapter unresponsive.
- SPI after a timeout. After an SPI transfer timed out, every later transfer came back one byte late until a restart. The controller is now reset after a timeout.
- Zero-length SPI transfer. Now returns
INVALID_PARAMETER. It returnedFW_UNHANDLED_ERROR(0x000C). - I2C target mode after large transfers. Starting I2C target mode after a large (bulk) SPI transfer no longer times out.
- Concurrent traffic with clock stretching. Commands sent while a long I2C clock stretch is in progress are no longer lost.
- I3C target table. Repeating RSTDAA followed by ENTDAA, SETDASA or SETAASA no longer adds a new table entry each cycle. Before, the table filled after about ten cycles and further address assignment failed.
- I3C external voltage check. Using an external supply on the I3C HV port now checks the HV voltage itself.
- Bootloader version query. Asking for the bootloader version now returns an empty string instead of unreadable bytes, and no longer resets the adapter.
Released 2026-09-30. For I3C HDR-DDR, use 4.6.1 instead of 4.6.0. Works with CosmicSDK 1.6.0 and PyCosmicSDK 1.6.0; no host change.
- HDR-DDR. A standard (SDR) command sent after an HDR-DDR read could time out after 1 s, and every I3C command after it then failed with
I3C_ENGINE_BUSYuntil a reset. On 4.6.0 this happened in about 90 % of DDR read → GETPID cycles. Measured on 4.6.1: 0 timeouts in 6000 cycles.
Known limitations
Section titled “Known limitations”- An IBI that arrives on the same clock as the start of a controller command is dropped, and that command times out. Seen once in 3000 cycles.
- Without a duty cycle set, some I2C frequencies round up: 550 kHz runs at 600 kHz. Setting any duty cycle gives the exact clock. Fixed in 4.6.2.
- An I2C frequency above 3 MHz (including 3.4 MHz) or below 11 Hz is accepted and reported as success, but the bus runs at about 11 Hz. On 4.5.0 and earlier the adapter could re-enumerate instead. Initialize I2C between 11 Hz and 3 MHz. Fixed in 4.6.2: the request returns
INVALID_PARAMETER. - SPI reads above about 20 MHz on a Supernova, or 25–50 MHz on a Pulsar (depends on wiring and the connected part), can come back shifted by one bit while reporting success. Writes are not affected. Keep reads that carry data at a lower clock, and put a time limit on flash busy (
WIP) polling. - A zero-length SPI transfer returns
FW_UNHANDLED_ERROR(0x000C). Fixed in 4.6.2. - In CAN-FD mode (Pulsar), large (bulk) transfers are not available. Normal I2C, SPI, UART and GPIO commands keep working.
- After large (bulk) transfers, starting I2C target mode can time out until the adapter is reset. Reset the adapter before starting target mode. Fixed in 4.6.2.
- SPI target mode can skip the first byte sent in the first transaction after a role change. Later transactions are correct.
Released 2026-09-29. Works with CosmicSDK 1.6.0 and PyCosmicSDK 1.6.0. Hosts that do not use the new features need no change. For HDR-DDR, install 4.6.1 instead.
Highlights
Section titled “Highlights”- I2C SCL duty cycle and exact frequency. Set the share of each SCL period held high, from 20 % to 80 % in 10 % steps. With a duty cycle set, the bus runs at an exact clock at or below the request, at most 0.29 % under between 1 kHz and 1.2 MHz (550 kHz runs at 549.8 kHz). The reply reports the frequency and duty cycle in effect.
- Clock stretching for I2C devices on the I3C port (Supernova). For SMBus and MCTP devices that stretch the clock. Runs at 10 kHz to 1 MHz with the SMBus 35 ms timeout; the actual rate depends on pull-ups and bus capacitance and is reported back. While it is on, I3C commands return
I3C_I2C_CLOCK_STRETCHING_ACTIVE. - SMBus reset and SDA pull-up on the I3C port (Supernova). Hold SCL low for 1 µs to 1 s (the SMBus/NVMe-MI reset), and set the SDA pull-up to AUTO (default), OFF or ON.
Behavior changes
Section titled “Behavior changes”- I2C transfers to an I3C dynamic address return an error. A legacy I2C transfer to the dynamic address of an I3C target now returns
INVALID_PARAMETER. Before, it knocked that target off the bus until it was power-cycled.
- Pins stop driving when the target voltage is off. With the I2C/SPI/UART/GPIO rail set to 0 V, SPI, UART and GPIO pins no longer drive an unpowered target. They resume, with the settings, when the rail comes back.
- Supernova I3C HV voltage accuracy. The I3C HV rail now lands within −7 to +2 mV of the request across 1200–3250 mV (1200 mV used to read 1075–1094 mV). Setting it takes about 550 ms.
Released 2026-09-24. A fix release on top of 4.5.0; no host change.
- SPI transfer length. A standard SPI transfer longer than 1024 bytes returns an error again, as in 4.4.0. Use the bulk transfer for larger payloads.
- I2C after arbitration lost. After an arbitration-lost error the next I2C command went unanswered, and the adapter restarted about 8 s later. The controller now recovers.
Known limitations
Section titled “Known limitations”- With the target voltage set to 0 V, GPIO outputs and SPI chip selects keep driving. Set outputs back to inputs before switching the rail off. Fixed in 4.6.0.
- A legacy I2C transfer to an enumerated I3C target’s dynamic address knocks that target off the bus until it is power-cycled. Fixed in 4.6.0 (the transfer returns an error).
- A standard (SDR) command sent after an HDR-DDR read can time out after 1 s, and every I3C command after it then fails with
I3C_ENGINE_BUSYuntil a reset. Rare on 4.5.1: about 2 in 300 DDR read → GETPID cycles. Fixed in 4.6.1.
Released 2026-09-16.
Highlights
Section titled “Highlights”- Transfers up to 32 KB. I2C and SPI transfers can carry up to 32,768 bytes on the Supernova and 16,384 on the Pulsar, and I3C up to 32,767 bytes per frame on the Supernova, using the adapter’s USB bulk endpoints. The SDK picks this path automatically.
- I2C and SPI target mode. The adapter can act as a device on the bus:
- I2C target: memory, EEPROM (page writes, write cycle, write protect) or register map.
- SPI target: memory, SPI flash (RDID, READ, FAST_READ, PAGE_PROGRAM, status, write enable) or register map.
- Release an interface. I2C and SPI can be released during a session, to switch between controller and target mode without a power cycle.
Behavior changes
Section titled “Behavior changes”- The adapter no longer stops answering all commands after SPI activity.
- An SPI transfer that runs over its time budget is answered only after it has stopped.
- Unknown commands return
FW_UNSUPPORTED_COMMANDinstead of stopping that interface.
Known limitations
Section titled “Known limitations”- I3C transfers are limited to 32,767 bytes per frame, counting the register address. A 32,768-byte I3C frame reports success but drops the last byte, and the next I3C transfers fail until the controller is re-initialized. This limit applies to every later version too.
4.4.0 and Earlier
Section titled “4.4.0 and Earlier”For notes on firmware 4.4.0 and earlier, contact support. Firmware 4.3.0 is the first version that uses the current bootloader.
Other products
Section titled “Other products”| Product | Firmware | Notes |
|---|---|---|
| Binho Atira | Latest on dfu.binho.io | Updated the same way as the Supernova and Pulsar: see Updating Firmware. |
| Binho Nova | See the Latest Firmware table | Release notes are not published for the Nova. |

