Ineda Systems, a company building out a new type of semiconductor for the internet of things, has raised $17 million in second-round funding. The startup, which is also pursuing a data center chip design, raised the money from Walden-Riverwood Ventures, Samsung Catalyst Fund, Qualcomm Ventures, IndusAge Partners and others, including Imagination, whose MIPS CPUs are used in Ineda’s products.
Like several other companies, Ineda is focused primarily on the wearable market for now, building a microprocessor designed to consume much less power. Unlike other architectures from ARM or Intel, its design is optimized to add increasing levels of performance in a system on a chip so companies can pick the lowest performing chip that will do the job in order to save power. Ineda calls this the hierarchical computing architecture, and it has built four different designs of its Dhanush wearable processing unit featuring the architecture.
The lowest level of the hierarchy is an always-on computing core that performs the basic needs and can integrate with a sensor platform. The second level handles simple applications with some accelerators for bare bones graphics, while the third level can handle rich applications and manage an operating system. The IO, file system, memory, display and other elements on a chip are shared between all the compute cores. The resulting systems on a chip come in four different classes designed for different devices that could range from a connected ring to a high-end smart watch, as listed below:
- Dhanush Advanced: Designed to include all the features required in a high-end wearable device – rich graphic and user interface – along with the capability to run a mobile class operating system such as Android.
- Dhanush Optima: This is a subset of the Dhanush Advanced and retains all the same features except the capability of running a mobile class operating system. It offers enough compute and memory footprint required to run mid-range wearable devices.
- Dhanush Micro: Designed for use in low-end smartwatches that have increased compute and memory footprint. This contains a sensor hub CPU subsystem that takes care of the always-on functionality of wearable devices.
- Dhanush Nano: Designed for simple wearable devices that require microcontroller-class compute and memory footprint.
Ineda isn’t just thinking about the wearable market. The act of sharing IO and memory between tiny low power cores is actually a problem in another area of computing — that of microservers used in the data center. These servers — championed by Seamicro, which was purchased by AMD, or Calxeda, which has shut down — are designed to be highly dense and optimized for many small workloads and consume less power than machines optimized for performance.
Ineda CEO Dasaradha Gude explained that the company is developing a processor for that market called Gemini that will slip into the PCI Express slot and handle converged memory and networking functions. That product is still in stealth mode and won’t even be fully designed until the third or fourth quarter of this year. Meanwhile, the wearables product could find its way into devices by the end of 2014. Building new hardware is tough, but now, when a market is just beginning to gain shape, is the time to make a go of it with a substantially new architecture.
Ineda says its chips can deliver a 30-day battery life, always-on voice recognition, a better integration with both sensors and radios and several other features that could help drive adoption in the nascent wearables market. Its executives are former engineers at AMD, Qualcomm and other big-name chip firms, while Sanjay Jha, the former head of Motorola Mobility and COO at Qualcomm, is the chairman of the board.