RK1820 AI Processor Chip - Ultra-High-Bandwidth Co-Processor for AI Applications
The RK1820 AI Processor Chip integrates ultra-high-bandwidth on-chip DRAM with a multi-core NPU optimized for 3B/7B LLM/VLM applications. Supporting mainstream model frameworks with universal API for straightforward development, this co-processor delivers exceptional AI computing performance when paired with host controllers like the RK3588.
Product Overview
RK1820 combines ultra-high-bandwidth on-chip DRAM with a multi-core NPU specifically optimized for 3B/7B LLM/VLM models. The chip supports mainstream model frameworks and provides a universal API for simplified development workflows. When integrated with host controllers such as the RK3588, the RK1820 creates a powerful AI computing platform ideal for edge computing applications.
Multi-modal AI capabilities enable diverse applications including speech recognition, video analytics, and long-context dialogue systems. This makes the RK1820 suitable for deployment across security, robotics, automotive, consumer electronics, office, education, home automation, and industrial sectors.
Technical Specifications
- Three 64-bit RISC-V cores with FPU support
- Each core features 32 KB instruction cache + 32 KB data cache + 128 KB L2 cache
- Integrated NPU supports mixed-precision operations (INT4/INT8/INT16/FP8/FP16/BF16)
- Computational power up to 20 TOPS (precision/model dependent)
- Main RISC-V core handles model scheduling, control, and I/O operations
- NPU dedicated to inference processing
- Integrated ultra-high-bandwidth on-chip DRAM
- PCIe 2.0 and USB 3.0 PHY for seamless connectivity
AI Performance Capabilities
Supports 3B/7B-scale LLM/VLM models running at the edge. Capable of deploying up to 7B large models on offline edge devices to enable advanced AI functions including video summarization, image recognition assistance, and multimodal perception.
When running 7B models, the RK1820 achieves output performance at the level of hundreds of tokens per second. Actual performance varies based on model architecture, quantization precision, I/O bandwidth, and host collaboration efficiency.
Application Areas
The RK1820 is engineered for edge computing scenarios across multiple industries including security systems, robotics, automotive applications, consumer electronics, office automation, education technology, home automation, and industrial automation.
Additional Product Inventory
| Part Number |
Package |
| HMC634LC4TR | 24-TFQFN |
| HMC442LC3BTR | 12-VFQFN |
| HMC519LC4TR | 24-TFCQFN |
| HMC753LP4E | 24-TFQFN |
| HMC506LP4ETR | 24-VFQFN |
| MAX2615ETA | 8-WFDFN |
| ADL5604ACPZ | 16-LFCSP-VQ |
| HMC452ST89ETR | TO-243AA |
| HMC390LP4ETR | 24-QFN |
| HMC386LP4ETR | 24-VFQFN |
| HMC373LP3ETR | 16-VFQFN |
| ADL5724ACPZN | 8-UFDFN |
| HMC326MS8GETR | 8-TSSOP |
| ADL5721ACPZN-R7 | 8-UFDFN |
| HMC625BLP5ETR | 32-VFQFN |
| ADA4961ACPZN-R7 | 24-WFQFN |
| ADRF6520ACPZ | 32-LFCSP |
| HMC326MS8GTR | 8-MSOPG |
| HMC902LP3ETR | 16-QFN |
| HMC507LP5ETR | 32-VFQFN |
| HMC952ALP5GETR | 24-VQFN |
| HMC441LP3E | 16-VFQFN |
| HMC383LC4 | 24-TFQFN |
| HMC907APM5E | 32-LFQFN |
| HMC341LC3B | 12-VFQFN |
| ADL8104ACPZN | 16-LFCSP-WQ |
| HMC441LH5 | 12-LCQFN |
| HMC460-SX | Die |
| HMC-APH462-SX | MODULE |
Frequently Asked Questions
Are your products original?
Yes, all products are original. Supplying new original imported components is our primary business focus.
Which Certificates do you have?
We are ISO 9001:2015 Certified Company and maintain membership with ERAI.
Can you support small quantity orders or samples? Is the sample free?
Yes, we support sample orders and small quantity purchases. Sample costs vary depending on your order requirements and project specifications.
How do you ship my order? Is it safe?
We utilize major express carriers including DHL, FedEx, UPS, TNT, and EMS. We can also accommodate your preferred shipping provider. All products are carefully packaged to ensure safety during transit, and we assume responsibility for any product damage to your order.
What is the lead time?
Stock parts are typically shipped within 5 working days. For non-stock items, we will confirm the lead time based on your specific order quantity.