Generate a synthesizable RTL model of a neural network in minutes
USING cloud based app NAMI
A fully configurable neural network model
WITH SUPPORT FOR DIFFERENT TYPES OF TOPOLOGIES (FNN, RNN, CNN) AND ACTIVATION FUNCTIONS
TARGET either FPGA or ASIC
With a flip of a button
INTRODUCING CLOUD BASED APP NAMI
Implementing an artificial neural network in RTL can be time consuming and requires advanced design expertise to meet the very stringent latency and throughput requirements, as well as power consumption, of a modern AI system. Siwave’s Neural Network RTL generator app NAMI allows AI engineers to effortlessly implement their AI models in hardware without any expertise in RTL design or knowledge of the target hardware. RTL design engineers can also use the tool to experiment with different neural network configurations without impacting their project schedule.
NAMI is accessible through a secure web portal and can generate an encrypted IP for either FPGA or ASIC implementation in minutes. Nami can also, optionally, generate simulation models and implementation scripts to facilitate the verification and integration of the neural network IP.
NAMI can now generate Intel Platform Designer compliant IP to further speed up the deployment of AI system on Intel FPGA. Click here for details.
Support for different types of neural network and unlimited number of layers
FNN (MLP, FC, etc.)
RNN (o2o, o2m, m2o, etc.)
– A choice of four types of activation functions
(Binary, Relu, HardTanh and Softmax)
– Configurable width for both data and weights
(fractional and integral parts)
– Full access to internal processing nodes for
monitoring and debugging
– Option to add a custom activation function
– Automatic throughput and latency adjustment with a manual override
– Plus many more features
for FPGA and ASIC
SiWave develops innovative IP solutions for the embedded market. We have been in business since 2003 and have successfully delivered numerous projects targeting a wide range of different industry standards. We typically sell our IPs to IC vendors as well as OEM, IDM, silicon vendors, etc. looking to develop their own SoC solutions.