ESP32-H21

What changed from the H2

The H21 is not a new architecture. It is the same basic proposition as the H2 — a 96 MHz single-core RISC-V part with Bluetooth LE and 802.15.4 and no Wi-Fi — with three targeted improvements aimed squarely at battery-powered mesh nodes.

An on-chip DC-DC converter. This is the most consequential change and the least visible on a specification sheet. Without it, the radio’s supply is derived through a linear regulator, and the difference between the battery voltage and the radio’s operating voltage is burned as heat. A switching converter recovers most of that. Espressif quotes roughly 8.2 mA during receive as a result.

For a mesh device that is often the dominant energy term. A Thread router or a Zigbee node that has to keep its receiver on to relay traffic for its neighbours spends most of its life in RX, not TX, and not asleep. Cutting receive current is worth far more to that device’s battery life than shaving microamps off deep sleep.

5 µA deep sleep, down from the H2’s 7 µA, with 9 µA in light sleep.

Up to 20 dBm transmit power. That is 100 mW, a substantial link-budget improvement over typical output levels. In practice it buys range through walls, floors and industrial clutter, and it makes marginal links reliable rather than intermittent — which in a mesh network reduces retransmissions and therefore, indirectly, saves energy again.

What is unchanged

Everything else is essentially H2-equivalent, and the limitations carry over:

• No Wi-Fi. Still requires a border router or gateway to reach an IP network.
• 19 GPIOs. A sensor node’s worth of pins, no more.
• No LP core — only LP peripherals. Nothing executes your code during deep sleep.
• 320 KB SRAM plus 128 KB ROM, with 4 KB of LP SRAM surviving sleep. External flash over SPI; no in-package flash.
• USB Serial/JTAG only — no USB OTG.
• Bluetooth is LE 5, certified against BLE 5.3. Not the 5.4 of the H4 and S31, so no LE Audio.

Hardware crypto matches the H2 — AES, SHA, RSA, ECC, HMAC and a TRNG — but the H21 does not carry the APM/TEE isolation found on the C6 and P4, and the published tables do not credit it with the dedicated ECDSA digital-signature peripheral the H2 has.

Software support

The H21 is supported by ESP-IDF, with preview support landing in v6.0 in March 2026. Espressif has confirmed ESP-Matter SDK support for Matter-over-Thread development and ESP-BLE-MESH for Bluetooth mesh, plus the standard BLE peripheral and advertising roles. Because it shares the H2’s architecture closely, most existing H2 code and component choices carry across.

H21 or H2?

If both were equally available, the H21 would be the obvious choice for any battery device — it is strictly better on power and range with no functional regression. Since they are not equally available, the decision is a scheduling one:

• Shipping in 2026 → H2. It is in production, priced ($1.39), and has a longevity commitment running from 2021.
• Designing for 2027 → design for the H21, and validate on H2 hardware in the meantime.
• Need two cores, more pins, or Bluetooth 5.4 → neither; look at the H4 .

Choose it when

• The device runs on a coin cell or a primary battery and every microamp is scrutinised
• It spends significant time in receive, as a Thread router or Zigbee relay would
• Range is marginal — thick walls, basements, industrial environments
• You are building Matter-over-Thread sensors or BLE mesh lighting at scale

Choose something else when

• You need parts now, in volume → ESP32-H2
• You need Wi-Fi → ESP32-C6
• You need more than 19 GPIOs, or LE Audio → ESP32-H4
• You need a TEE or PSA certification → ESP32-C6