New solutions to Edge Computing: Enabling Smarter Edge at ultimate energy efficiency

Dolphin Design is now accelerating the development of ML/AI IPs to accelerate the time-to-market of our partners addressing diverse market verticals such as voice-controlled systems, access control and any edge-IoT applications where processing performance and power efficiency are required simultaneously.
Dolphin Design’s flexible offering ranges from power efficient, core agnostic MCU sub-systems platform (Chameleon) to highly configurable DSP platforms (Panther) up to Neural processing unit (Raptor) for high performance AI applications detailed below:

– Chameleon is a configurable MCU sub-system embedding several standard peripherals, an autonomous DMA, a fined-grained power management unit, a low latency interconnect, an event manager and also a tiny ML accelerator for specific applications. Thanks to these specific features, extremely low power consumption can be achieved in both deep sleep and active modes with a high score in the industry reference EEMBC uLPMArk (Core and Peripheral Profile, 500 & 120 respectively) for an implementation in a 22nm advanced process node. Chameleon platform comes with an RTL configurator tool as well with a Software Design Kit (SDK) which embeds all needed elements to easily port an existing SW.

– As a complementary offer to Chameleon, Panther is a highly configurable DSP platform based on up to 16 processing elements and a ML/AI accelerator. The platform embeds an event manager as well as a low latency interconnect to maximize the energy efficiency. The platform comes with an RTL configurator and a tool chain. Panther can enable >100 GOPS computing performance with an efficiency of 1TOPS/W (@10GOPS). First deployments will happen in Q3 2021.

– For high performance AI applications, Raptor is a Neural Processing Unit specialized in vision processing algorithm, which includes a host core, a DMA, and up to 128 processing elements. Raptor can enable more than 60 GOPS computing performance and an efficiency of 2200GOPS/W (@16GOPS) in 28nm FD-SOI technology. It will be fully available Q1 2022.

These 3 digital platforms are, integrated inside a test chip, open the possibility to address several IoT market verticals with leading edge power versus performance ratio. Several targeted applications will be detailed in our presentation.