Alp SDK v0.8 — E1M-AEN801 silicon bring-up on Alif Ensemble E8
A second SoM family now carries silicon evidence: 15 of 17 peripheral apps PASS on real E8 hardware, with real NPU inference from MRAM on Ethos-U85, dual-core HE↔HP RPC, SE-CryptoCell compute, and the CC3501E Wi-Fi/BLE bridge. v0.8.1 follows with a repo-wide silicon-accuracy doc sweep.
Alp SDK v0.8 is the first full bench bring-up of the E1M-AEN801 (Alif Ensemble E8, Cortex-M55) on real silicon — making it the second SoM family, after the E1M-X V2N, to carry hardware evidence rather than build-only validation.
The peripheral matrix is the headline: 15 of 17 aen-* example apps PASS on real E8, each flashed over J-Link direct-MRAM (Flow D) and booted from MRAM — GPIO, UART, PWM, SPI, I²C with an EEPROM on the bus, ADC and DAC, the watchdog, PDM microphones capturing live PCM, and I²S tone-out. The two PARTIAL cases — the quadrature encoder and the SD card — are hardware-gated on the bench, not code or flashing bugs.
A real person_detect MobileNet runs on the Ethos-U85 NPU with the model resident in MRAM slot0, and the two M55 cores talk to each other end-to-end: an Arm MHUv2 doorbell, 64 of 64 shared-memory round-trips, and an OpenAMP RPMsg pingpong at 16 of 16. SE-CryptoCell AES/SHA/AEAD compute is bench-validated and on by default, and the CC3501E Wi-Fi 6 + BLE 5.4 bridge runs real Wi-Fi and BLE scans over a hardware SS0 chip-select.
The generalizable lesson came out of the Ethernet bring-up: any DMA-master block on the E8 M55 — the GMAC, the Ethos-U NPU, the SD host — needs its DMA-visible buffers in global on-chip SRAM, never the default DTCM, or the transfers silently starve. Ethernet now pulls a DHCP lease end-to-end once its descriptor rings move to SRAM0.
Status stays honest. Two families now carry silicon evidence — the V2N GD32 supervisor bridge and the AEN801 peripheral matrix — while i.MX 93, V2M/DEEPX, and the AEN30x–70x parts remain [UNTESTED] on real hardware. Wi-Fi and BLE are not yet proven concurrent on the bridge; that path is conf-gated, not demonstrated. The follow-on v0.8.1 is a repo-wide accuracy sweep that corrects the documentation to exactly this reality, so nothing on the page claims more than the bench has shown.
