Allwinner T536 SoM targets edge AI
- Boardcon launched the PICOT536 system-on-module and EMT536 carrier board on May 4, built around Allwinner’s T536 industrial edge-AI processor. (cnx-software.com) - The key spec is a 2 TOPS NPU beside four Cortex-A55 cores, dual RISC-V controllers, inline ECC memory, and wide industrial I/O. (cnx-software.com) - It matters because T536 designs are moving from chip announcements into shippable modules for cameras, HMIs, robots, and local inference boxes. (cnx-software.com)
A system-on-module is basically the shortcut version of board design. You get the processor, memory, storage, power basics, and a big pile of I/O on one c(cnx-software.com) running a model while also talking to cameras, displays, Ethernet, motors, and industrial buses. Boardcon’s new PICOT536 and its matching EMT536 dev board push Allwinner’s T536 into exactly that lane. (cnx-software.com) ### What actually launched? Boardcon launched the PICOT536 SoM and the EMT(cnx-software.com)t it, with storage, display outputs, and expansion already broken out. Boardcon is pitching the combo at industrial HMIs, machine vision, robotics, and other edge-computing jobs. (cnx-software.com) ### What is inside the T536? The Allwinner T536 is a mixed-architecture industrial SoC. It has four Arm Cortex-A55 CPU cores up to 1.6 GHz for the Linux side, plus XuanTie E907 and E902 (cnx-software.com)ck, another handles deterministic chores like sensor handling or control loops. (cnx-software.com) ### Why does that mix matter? A lot of edge devices are doing two jobs at once. A smart camera might need to run object detection, encode video, and still respond instantly (cnx-software.com)etup is useful because it avoids forcing one general-purpose CPU cluster to do everything badly. (boardcon.com) ### What does the module itself give you? The PICOT536 comes with 2 GB, 4 GB, or 8 GB of LPDDR4/LPDDR4X memory with inline ECC, plus 8 GB to 64 GB of eMMC. It uses a 314-pin MXM 3.0 edge connector and exposes camera lanes, LVDS and MIPI DSI display outputs, (cnx-software.com)sit in a real machine, not just a demo box. (cnx-software.com) ### Is this really about AI? Yes, but not the cloud-datacenter kind. The NPU supports INT8, UINT8, and INT16 inference, and Boardcon explicitly frames it around object detection, defect inspection, and voice commands on-device(boardcon.com)fewer headaches when a product has to keep working even with spotty connectivity. Think factory cameras, kiosks, charging stations, or embedded vision nodes. (cnx-software.com) ### What makes it industrial instead of hobbyist? Two things stand out. First, temperature range — Boardcon lists an industrial option from -40(cnx-software.com)ort around things like CAN and RS485 in the software stack. Allwinner also positions the T536 family around industrial and power-sector use, with ECC support and lots of field-bus-friendly connectivity. (cnx-software.com) ### Why is this showing up now? The bigger story is that the T536 is moving from “interesting(cnx-software.com)ally matters more than the raw silicon announcement, because developers buy modules and dev kits long before they commit to a custom board spin. (cnx-software.com) ### Bottom line? This launch does not change the edge-AI market overnight. But it does make Allwinner’s T536 more real. The useful part is the combination — modest 2 TOPS AI, lots of industrial I/O, EC(cnx-software.com) boxes, that is often the difference between a chip you admire and a product you can finish. (cnx-software.com)