The fundamentals of GAP, the IoT application processors
Enables AI and DSP applications on battery-powered devices
GAP processors bring entirely new levels of computing capability into the realm of MCU-class energy consumption. GAP8 and GAP9 power new types of devices that combine ultra-low power consumption with sophisticated signal processing and neural network algorithms.
Flexible and programmable
The GAP RISC-V based processors are fully programmable in C/C++ giving you the ability to optimize the latest algorithms. The GAP SDK includes all the tools necessary to speed application development including a fully automated neural network toolchain from Google Tensorflow. The unique GAP AutoTiler code generator automatically optimizes the flow of data across the chip.
Dynamically adapts power consumption to the needs of the application
GAP processors incorporate dynamic voltage and frequency scaling and automatic clock gating keeping awake only the elements of the component necessary to service the current workload. Low standby power consumption and ultra-fast wake-up and power state transitions minimize energy usage across sleep, acquisition, processing, and communication states.
GAP8 | GAP9 | |
Processing power | 22.65 GOPS | 150.8 GOPS |
Power efficiency | 4.24 mW/GOP | 0.33 mW/GOP |
Memory | ||
| L1 | 80 kB | 128 kB |
| Ram | 512 kB | 1.5 MB |
| Flash | None | 2 MB |
| External | QSPI/ HyperBus | 2x QSPI/OCTO-SPI/HyperBus/SDIO |
MAX Frequency | ||
| FC* | 250 | 250 |
| Cluster** | 175 | 400 |
Fixed Point | 8, 16, 32-bit | 8, 16, 24, 32-bit |
Floating Point | None | 8/16/32-bit |
Sound Interface | 2 Rx-Only I2S interfaces | 3 master/slave SAI full duplex, I2S and TDM 4/8/16 ch capable |
Camera Interface | 8-bit CPI (Camera Parallel Interface) | 8-bit CPI, 2-lane CSI-2 |
Package type | aQFN 88 7x7mm | BGA 6x6mm – WLCSP 3.5x 3.5mm |
Availability | In production | Samples in 2020. In production 2021 |
*FC (Fabric Controller) is the main system controller and resembles a standard MCU; it can delegate compute-intensive tasks to the Cluster.
**GAP includes a multicore compute cluster with a shared memory architecture and hardware thread synchronization. The cluster enables highly efficient, parallel implementation of algorithms giving almost optimal linear speedup.
GAP8
GAP8 allows industrial and consumer product manufacturers to integrate signal processing, artificial intelligence and advanced classification into new classes of battery-operated wireless edge devices for IoT applications including counting people and objects, human detection, head pose detection, sound analysis, face identification, speech recognition, activity classification, etc.
GAP9
GAP9 extends the accessible market of the GAP family to support more complex applications at even lower power consumption. Through a combination of a state-of-the-art 22nm FD-SOI semiconductor process and architecture improvements such as increased memory bandwidth, data compression, and trans-precision floating point support, GAP9 reduces energy consumption by 5 times while enabling inference on neural networks that are 10 times larger. GAP9 delivers a significant performance upgrade while continuing to support the requirements of battery-powered sensors and wearable devices.