Kaviaz Technology Kaviaz Technology
An engineer reviews a holographic system diagram above the desk, showing interconnect fabric, memory controller, PHY and NPU compute tiles connected as a unified block diagram
Semiconductor IP

Openedges

ORBIT memory subsystem: interconnect, controller and PHY from one source

  • Interconnect
  • DDR Controller
  • DDR PHY
  • LPDDR5X
  • HBM3
  • NPU

Openedges is a South Korean (KOSDAQ-listed) memory-subsystem IP licensor positioned as a leader in smart computing, with 40+ customers and 80+ IPs deployed worldwide, emphasizing memory-AI system co-design. It supplies the complete ORBIT memory subsystem IP, combining on-chip interconnect, DDR memory controller and DDR PHY from a single vendor and removing the overhead of integrating multiple IP suppliers. The ENLIGHT™ mixed-precision NPU IP is also available.

Visit website
Products & capabilities

Openedges portfolio

On-chip interconnect

OIC — On-chip Interconnect IP

Uses a proprietary HXI protocol to reduce wire count, plus HyperPath™ for path-based clock control via a side-band channel, claiming a 2x performance gain with HyperPath Express. Runs above 1GHz at 28nm with backbone idle power under 200µW, and bridges physically distant clock domains via LDA technology.

Visit website
Memory controller

OMC — DDR Memory Controller IP

JEDEC-compliant DDR memory controllers covering LPDDR6/5x/5/4x/4, DDR4/3, GDDR6 and HBM, with proprietary out-of-order scheduling achieving over 90% DRAM utilization. A single OMC instance drives dual PHYs to double channel bandwidth, with AES memory encryption, inline ECC and ISO 26262 safety certification.

  • LPDDR5X / 5 / 4X / 4
  • DDR5 / 4 / 3
  • GDDR6
  • HBM3
Visit website
Memory PHY

OPHY — DDR PHY IP

DDR PHY IP validated together with OMC, supporting LPDDR6/5X/5/4X/4, DDR5/4/3, GDDR6 and HBM3 with JEDEC and DFI compliance — LPDDR5X reaching up to 10,667 Mbps. Process nodes from 4nm to 28nm, with an embedded programmable state-machine (PSM) microcontroller.

  • LPDDR5X / 5 / 4X / 4
  • DDR5
  • GDDR6
  • HBM3
Visit website
AI acceleration

ENLIGHT™ NPU IP

Available as ENLIGHT Classic (4/8-bit mixed-precision, edge-focused) and ENLIGHT Pro (scalable inference NPU for automotive and camera applications). ENLIGHT Pro delivers 4096 MACs/cycle at 8-bit integer, with the product range spanning 8 TOPS to hundreds of TOPS across single/dual/quad-core configurations.

Visit website

Want to know more about Openedges?

We can arrange a vendor technical briefing, an evaluation licence and a deployment plan to get your team productive quickly.