Skip to content

Day one: a dead cooker and a plan

On 28 or 29 October my Anova Precision Cooker Pro started a firmware update from the app, got to 75 %, and stayed there. After a night plugged in and a power cycle it was dead: no beep, no display, no LEDs. The cooker had worked perfectly until then. Support's advice was to leave it plugged in overnight; that changed nothing. On 1 November the answer was that the unit is out of warranty, and 10 % off a new one.

So on 1 November I opened it, and this repository started.

What's inside

The head comes apart with a cut around the glued glass and four plastic screws. The main board hinges out on a 30-way flat cable and a speaker lead. On it, three chips matter:

  • U1, a TI CC3220S: the Wi-Fi microcontroller. It has no internal flash; it boots from
  • U15/U16, an 8 MiB SPI flash (XMC XM25QH64A), which holds the firmware and the file system.
  • U7, a TI CC2640R2F: the Bluetooth chip, with its own flash inside.

Two more chips are unidentified for now, and the power board hasn't been looked at yet. The full picture is in the hardware overview; every chip in the component list carries a confidence label and its source.

The puzzle

Here's the part that makes this interesting rather than just annoying. The CC3220's maker documents an over-the-air update mechanism with a rollback: the previous firmware is kept until the new one has proven itself. The flash on this board has room for both. So "completely dead after an interrupted update" shouldn't be a possible outcome, and the fact that it happened is itself a clue. Either the firmware doesn't use that mechanism, or the update touched something it doesn't cover (the Bluetooth chip, the file system itself), or the problem isn't in the Wi-Fi chip at all. A dump of the flash will say which. That's the first real milestone.

The rules

Before any of that, rules, because the brick came from an interrupted write and the recovery must never cause a second one:

  1. Nothing gets written to any flash until a backup exists that was read twice with the same hash.
  2. The second, working cooker in the house is read-only. It's the reference, not a donor.
  3. The power board carries mains. Nothing on it is probed while plugged in.
  4. Every claim gets a confidence label and a source.

The plan, such as it is

There isn't one, deliberately. We don't know what's in front of us, so the next stage is open exploration: tools arrive over the coming weeks (a logic analyser, a flash clip, a microscope, a known-good CC3220 development board), and each one gets used to learn as much as it can. The route gets chosen when the findings make it obvious. Three routes are on the table: fix the original board, replace it with a used one, or build our own around an ESP32. The last one is the committed fallback. Anova is not counted on.

Tools bought on day one: about €545. More than the discount is worth, by some margin. That's fine.

What's next

Four things cost nothing and can happen before any parcel arrives: a Bluetooth and Wi-Fi scan of the dead unit (does anything on the board still talk?), continuity mapping of the connectors with a multimeter, reading up on how Anova's updates actually work, and watching the working unit's power-on sequence with a stopwatch. Then, on Tuesday, the first parcel.

Everything learned goes in the timeline the moment it's learned.