Raysen Electronic Components Distributor

Hong Kong electronic component distributor with a searchable catalog of 110,000+ Texas Instruments part numbers, covering specifications, datasheets and stock status.

Core Strengths

Industry-Leading Chips

Industry-leading semiconductor brands as core components, delivering advanced performance, stability and compatibility.

TI Power Management

Multiphase smart power stages, integrated GaN power stages and hot-swap protection devices.

ADI Signal Chain

μModule regulators, programmable power sequencers and isolation/signal conditioning devices.

Two AI Modules

Two AI server modules successfully developed for edge computing and data center acceleration.

Technical Q&A

Key technical Q&A on AI data center power management and architecture

What new power management chips did Texas Instruments launch in 2026 for AI data centers?
TI introduced the TPS1685 (a 48V integrated hot-swap eFuse with power path protection) and the LMG3650R035/R025/R070 family (integrated GaN power stages in industry-standard TOLL packaging). The TPS1685 supports power levels above 6 kW, while the GaN power stages achieve >98% efficiency and >100 W/in3 power density.
What are the core characteristics of the 800 VDC power architecture co-developed by TI and NVIDIA?
The architecture reduces the conversion path from 800V to GPU core voltage to just two stages: an 800V-to-6V isolated bus converter (97.6% peak efficiency, >2000 W/in3 power density), followed by a 6V-to-sub-1V multiphase step-down solution. This substantially simplifies the power distribution hierarchy compared with conventional architectures.
What specific components comprise TI's 800 VDC architecture?
It includes an 800V hot-swap controller, an 800V-to-6V DC/DC converter with integrated GaN power stages, a 6V-to-sub-1V multiphase step-down converter, a 30 kW 800V high-power-density AC/DC PSU, and an 800V capacitor bank using EDLC supercapacitor cells (40 W/in3 power density).
How does the TPS1685 eFuse improve upon conventional hot-swap controllers?
The TPS1685 is a 48V stackable integrated hot-swap eFuse with power path protection. Compared with existing hot-swap controllers on the market, it simplifies design and halves solution size, while scaling to meet high-power handling requirements above 6 kW.
What are the key specifications of TI's new LMG3650 GaN power stage family?
The LMG3650R035/R025/R070 integrates a high-performance gate driver with a 650V GaN FET, achieving >98% efficiency and >100 W/in3 power density. Integrated protection features include overcurrent protection, short-circuit protection, and overtemperature protection, making it suitable for server power supplies and other AC/DC applications.
How does ADI's MAX20816 multiphase controller support AI accelerator power delivery?
The MAX20816 is built on a dual-loop flexible architecture supporting up to 16 configurable phases distributed across two independent rails. It achieves fast, precise voltage control through PWMVID and AVSBus digital interfaces, supports coupled inductors and smart power stages, and scales to 32 or 48 phases with current capability up to 2500A.
What telemetry and system reliability features does the MAX20816 offer?
It integrates PMBus telemetry for real-time monitoring of voltage, current, input power, and power-stage temperature. Built-in non-volatile memory stores preset or custom configurations, while fault logging and protection features help prevent critical failures, particularly in high-power-density environments.
What is the strategic purpose behind ADI's acquisition of Empower Semiconductor?
ADI acquired Empower for $1.5 billion to strengthen its high-density power electronics position for AI applications. Empower developed integrated voltage regulators (IVRs) and silicon-based capacitor technology that enable power conversion closer to the processor, shortening the power path and reducing losses. ADI's CEO noted that energy delivery has become the primary bottleneck for AI system scaling.
What is the core component for power path protection in AI servers?
The TI TPS1685 is representative: a 48V integrated hot-swap eFuse with power path protection. It enables designers to meet high-power handling requirements through scalable devices while simplifying design and reducing size. It is a key component for protecting CPUs, GPUs, and AI accelerators in 48V power architectures.
Why are AI data centers transitioning from 12V to 48V power architectures?
As AI workloads surge, data center designers are moving to 48V power architectures to improve efficiency and scalability, better supporting high-power components such as CPUs, GPUs, and AI hardware accelerators. TI's 48V stackable hot-swap eFuse is designed precisely to meet power handling requirements above 6 kW.
Which GaN devices are used in TI's 1 kW server PSU reference design?
The TIDA-010062 reference design uses LMG341x GaN FETs with integrated drivers for the continuous conduction mode totem-pole bridgeless PFC stage, and LMG3422 GaN FETs for the half-bridge LLC isolated DC/DC stage. The PFC stage achieves >98.8% peak efficiency and the LLC stage >98.3%, meeting 80 Plus Titanium standards.
What is the digital control core of this 1 kW reference design?
A C2000 Piccolo microcontroller controls both power stages (PFC and LLC), enabling a compact digitally controlled design suitable for server power supply units (PSUs) and telecom power rectifier applications.
What is Arm's core assessment of CPU demand in the Agentic AI era?
Arm CEO Rene Haas noted that under the same power constraints, future data centers may require more than four times the number of CPU cores compared with today. Beyond generating answers, Agentic AI must continuously execute tasks, invoke tools, and manage workflows, with the resulting scheduling, distribution, and coordination loads increasingly borne by CPUs.
What is the core competitiveness of Arm's AGI CPU?
Energy efficiency is central. The product delivers approximately twice the performance per watt of comparable x86 systems. Under air cooling, 36 kW supports 8160 cores and >180 TB of memory; under liquid cooling, 200 kW supports >45,000 cores and >1 PB of memory. Large-scale adoption of Arm-based CPUs could reduce power demand by approximately 10 GW.
What are the key electrical specifications of the TPS1685 hot-swap eFuse?
According to TI's official product data, the TPS1685 is a stackable integrated hot-swap eFuse with a 9V-to-80V input range and 3.5mΩ typical on-resistance. Its continuous absolute maximum voltage is 92V, the configurable current limit spans 2A to 20A, typical quiescent current is 2.2mA, and the operating temperature range is −40°C to 125°C (datasheet Rev. D).
How does the TPS1685 respond to overcurrent, overvoltage and short-circuit conditions?
Overcurrent is handled with a circuit-breaker response; fault response is configurable as auto-retry or latch-off; overvoltage triggers a cut-off. The device integrates overcurrent protection, overvoltage protection, short-circuit protection and thermal shutdown, and provides a current monitor output, fault output and a power-good signal, with adjustable soft start.
Which GaN power stage part numbers does TI recommend for rack power, and how should they be selected?
TI's rack power page highlights three devices: the LMG3100R017 (100V, 1.7mΩ, integrated driver), the LMG3522R030 (650V, 30mΩ, integrated driver with protection and temperature reporting), and the LMG3650R025 (650V, 25mΩ, TOLL package, integrated driver with protection and zero-voltage detection). Choose the LMG3100R017 for low-voltage high-current stages, the LMG3522R030 when temperature reporting is needed, and the LMG3650R025 when zero-voltage detection and TOLL packaging are required.
What are the key specifications and integrated features of the LMG3650R025?
It offers 650V VDS, 25mΩ RDS(on) and 30A maximum drain current in a TOLL package with bottom-side cooling. Integrated features include cycle-by-cycle overcurrent protection, latched overcurrent protection, slew-rate control and zero-voltage detection (ZVD), with an operating temperature range of −40°C to 150°C (datasheet Rev. A).
What control strategy and key part numbers does the PMP23537 reference design use?
This is a 3.6kW single-phase totem-pole bridgeless PFC reference design targeting M-CRPS server power supplies. It operates in continuous conduction mode (CCM) at the AC peak and in triangular conduction mode (TCM) with zero-voltage switching (ZVS) at low AC line, improving light-load efficiency by up to 2%; a baby boost stage extends holdup time and shrinks the bulk capacitor. Key devices include the LMG3427R030 (600V, 30mΩ GaN with integrated driver, protection and zero-current detection), the TMS320F280039C control MCU (C2000 real-time MCU, 120MHz, 384KB flash, CAN-FD), and the AMC1306M05 for 0.5%-accuracy energy metering.
Which cloud providers have adopted Arm's AGI CPU?
Oracle Cloud Infrastructure (OCI) and ByteDance have adopted it. In addition, data center CPU products including Google Axion, AWS Graviton, and NVIDIA Grace and Vera are all designed on Arm architecture.
How did Broadcom's AI semiconductor business perform in fiscal 2026?
Broadcom reported quarterly AI semiconductor revenue of $16.7 billion, up 221% year-over-year and representing 56% of total revenue. Custom chips accounted for 73% of AI semiconductor revenue. Google's latest TPU and OpenAI's Jalape?o chip have begun shipping, with custom chips for Meta starting shipments this quarter.
What is Broadcom's forward outlook for AI semiconductor revenue?
Broadcom raised its fiscal 2026 AI semiconductor revenue guidance to $58 billion (up 186% year-over-year), has secured supply to support a doubling to $115 billion in fiscal 2027, and potentially another doubling to $230 billion in fiscal 2028. Jalape?o is expected to reach approximately 1.3 GW of deployment in 2027.
What is the positioning of Google TPU in WSC (Warehouse-Scale Computer) environments?
TPU emphasizes a robust software stack and programmability, with the XLA compiler performing key optimizations such as fusion. The VCU (Video Coding Unit) achieves breakthroughs in compute capacity, supporting new capabilities such as multi-output transcoding. AV1 encoding requires 5–10x the compute resources of VP9, and the VCU design space offers a broader range of trade-offs.
What is the core argument of Huawei's "τ scaling"?
The central fact of τ scaling is that in large AI clusters, more than 80% of energy consumption is attributable to data transmission, and more than 70% of system cost is allocated to data storage. Reducing data transmission time—chip-to-chip, rack-to-rack, and within-package—is at least as important as reducing compute time.
How does the Unified Bus reduce AI system latency?
The Unified Bus replaces traditional multi-protocol stacks (PCIe → NVLink/proprietary → Ethernet/InfiniBand) with a single protocol, enabling fully symmetric memory semantics exposure. End-to-end remote access latency drops from tens of microseconds at TCP/IP-level protocols to approximately 100 nanoseconds, reducing system τ by roughly 500x.
What problem does the Hi-ONE near-package optical engine solve?
When per-chip bandwidth reaches several Tb/s, copper cabling becomes physically infeasible due to SerDes capacity limits, wiring volume, and insufficient thermal dissipation. Hi-ONE provides 8 Tb/s per module, shortening SerDes transmission distance from approximately 100 cm to about 5 cm, while extending reach from less than 1 meter to 100 meters.
How does Google's IPU architecture separate the server control plane from the data plane?
The IPU separates the server control plane (monitoring, virtualization, container lifecycle, network and storage I/O) from customer workloads onto different cores. Control plane overhead can spike to more than 50% of total compute capacity. The IPU includes a PCIe switch and hardware-accelerated virtualization features such as flexible physical-to-virtual mapping, live migration, and tenant isolation.
What role do smartNICs play in WSC networks?
SmartNICs must satisfy requirements including high throughput, low latency, low tail latency, traffic shaping, OS bypass, line-rate encryption, and debuggability, while maintaining low host CPU utilization. Approaches include onload (host CPU provides low latency) and offload (the NIC handles packets and encryption). Google's IPU, which separates the control plane from the data plane, is a representative implementation.
Why must AI accelerators inevitably move toward 3D folding?
Geometry dictates this trend. In conventional 2.5D packaging, compute capacity scales with N2 (area), but memory bandwidth, interconnect, and power delivery scale only with N (perimeter). All signals and current must travel along the chip edge to reach internal compute resources, constituting a fundamental bottleneck.
What are the core characteristics of the NVIDIA RTX Spark platform?
This PC compute platform is powered by a MediaTek-customized Grace CPU based on Arm architecture. NVIDIA CEO Jensen Huang stated that PCs will shift from "humans directly running applications" to "compute platforms that execute tasks on behalf of agents." With the NVFP4 model compression format, future PCs will be able to run AI models with up to 100 billion parameters.
What fundamental transformation is underway in AI data center power delivery architecture?
A shift from traditional power distribution architectures to high-voltage direct current (HVDC), with liquid cooling becoming standard. The AI surge is driving data center infrastructure comprehensively toward high-efficiency liquid cooling and HVDC power architectures, with innovations in active and passive components at the chip level and packaging optimization serving as the underlying keys to supporting extreme power density.
What is the specific "two-stage conversion" path in TI's 800 VDC architecture?
Stage one: an 800V-to-6V isolated bus converter with integrated GaN (97.6% peak efficiency, >2000 W/in3). Stage two: a 6V-to-sub-1V multiphase step-down converter with high current density and a dual-phase power stage design. Only two steps are required to bring 800V down to GPU core voltage.
What are the capabilities and core controller of the PMP23377 battery backup unit (BBU) reference design?
This is a 1.4kW four-switch buck-boost DC/DC reference design targeting OCP Open Rack V3 Power BBU specifications (50V BBU). The converter transitions smoothly between buck, buck-boost and boost modes depending on the input-to-output voltage relationship, uses peak current mode control for fast load transient response, achieves >98.5% peak efficiency, and supports bidirectional conversion with average input or output current monitoring and limiting. The core controller is the LM51770 (a 78V wide-VIN bidirectional four-switch buck-boost controller), which eliminates the need for an external FET driver; each phase delivers 1.4kW and multiple phases can be paralleled.
How do TI GaN power stages help meet 80 Plus Titanium, M-CRPS and OCP ORv3 specifications?
According to TI, the integrated driver reduces circuit parasitics and shrinks design footprint, while integrated overcurrent and overtemperature protection enable power supply units (PSUs), battery backup units (BBUs) and intermediate bus converter subsystems to meet M-CRPS and OCP ORv3 specifications. Combined with C2000 real-time MCUs and the associated reference designs, 80 Plus Titanium efficiency levels are achievable. TI also published the white paper "Achieving GaN Products With Lifetime Reliability" (SNOAA68), describing how its GaN products achieve lifetime reliability through an in-house program spanning epitaxial growth, application reliability validation and support for new industry standards.
How does TI describe the evolution of AI data center rack power and thermal management?
Patrick Zeng, general manager of data center thermal management at TI, notes that a typical rack consumed about 5kW twenty-five years ago, while today's racks can draw more than 100kW, and within a few years some could exceed a megawatt. Nearly every watt a server draws is converted back into heat. As rack power moved beyond 20kW to 40kW, air alone can no longer remove heat quickly enough, so engineers are combining air and liquid cooling, with a cold plate sitting directly against the chip so that liquid carries heat away. When a GPU overheats it cannot produce tokens as quickly, so end users experience slower responses and operator costs rise.
Which TI devices are needed to build a liquid cooling coolant distribution unit (CDU)?
According to TI, motor drivers control the pumps and fans inside the CDU with built-in protection against failure; valves open or close based on real-time demand to regulate coolant flow; sensors continuously monitor flow, temperature, humidity and pressure to catch the earliest signs of a leak or a failing component; and microcontrollers process sensor data in real time and adjust the system accordingly. TI's portfolio spans the full coolant path, from the HVAC systems that cool an entire facility, to the CDU that feeds the racks, down to the individual server tray. Reliability is always the number one consideration: a single failed pump or misread sensor can take an entire rack of AI computing offline.
What does TI see as the next frontier in data center thermal management?
TI notes that the next frontier of reliability in data center cooling combines multiple sensors through edge AI and sensor fusion. Today's systems often rely on a single sensor to catch a leak; tomorrow's systems will analyze a web of real-time signals - motor current, compressor speed, humidity, pressure and temperature - to recognize subtle patterns that no single sensor can detect. Correlating those signals is too complex for traditional rule-based software, while edge AI provides the unique advantage of fast response time and low latency.

Applications

01

Edge Computing

02

Intelligent Inference

03

Data Center Acceleration

About Us

Company Profile

Raysen Industry Hong Kong Limited is a technology company headquartered in Hong Kong, specializing in the research, development, and manufacturing of servers and AI-related server modules. Serving global markets, the company delivers high-performance, high-reliability hardware solutions for data centers, AI computing, and industrial applications.

Core Business

Our core business focuses on the industrial R&D and manufacturing of AI-related servers and functional modules. We adopt industry-leading semiconductor brands as core components to ensure industry-leading performance, stability, and compatibility.

For power management and signal chain design, we adopt Texas Instruments' multiphase smart power stages, integrated GaN power stages, and hot-swap protection devices, together with Analog Devices' highly integrated μModule regulators, programmable power sequencers, and isolation/signal conditioning devices. This delivers significant advantages in power conversion efficiency, transient response, signal integrity, and system stability.

R&D Capabilities

With an experienced and technically strong R&D team, we continue to invest in proprietary innovation. We have successfully developed two AI server modules, widely applicable to edge computing, intelligent inference, and data center acceleration. These modules use Texas Instruments chips and deliver outstanding performance in power management, signal chain, and system stability.

Industry Landscape

The AI server market is shaped by three major categories of players: brand-name system manufacturers such as Dell PowerEdge, HPE ProLiant, and Super Micro; ODM giants including Foxconn, Quanta, Wistron, Inventec, and Pegatron; and AI chip ecosystem vendors such as NVIDIA Vera Rubin, Google TPU, and Inspur. Raysen Industry Hong Kong Limited positions itself within this ecosystem as a focused developer and manufacturer of AI server modules and functional hardware, supplying the upstream supply chain that powers these systems.

Vision

Looking ahead, Raysen Industry Hong Kong Limited will continue to deepen its focus on AI servers and modules. Driven by technological innovation and built on reliable quality, we will collaborate with global partners to drive the advancement of AI infrastructure.

  • Headquartered in Hong Kong, serving global markets
  • TI multiphase smart power stages and integrated GaN power stages
  • ADI μModule regulators and programmable power sequencers
  • Two AI server modules successfully developed
  • For edge computing, intelligent inference and data center acceleration

Products

AI Server Modules

Two modules successfully developed, using Texas Instruments chips with outstanding power management, signal chain and system stability.

Power Management Design

TI multiphase smart power stages, integrated GaN power stages and hot-swap protection devices.

Signal Chain Design

ADI highly integrated μModule regulators, programmable power sequencers and isolation/signal conditioning devices.

Industrial R&D and Manufacturing

Covering industrial R&D and manufacturing of AI-related servers and functional modules, ensuring performance, stability and compatibility.

In-stock parts

Fully specified devices, ready for datasheet and pricing lookup

Search parts

Technology & R&D

An experienced R&D team with continuous investment in proprietary innovation. Driven by technological innovation, built on reliable quality.

TI

Power & Power Stages

ADI

Regulation & Signal Conditioning

AI

AI Server Modules

Contact Us

We welcome inquiries from global partners and customers and look forward to advancing AI infrastructure together.

Company Raysen Industry Hong Kong Limited
Email sahat@raysen.hk
Address Reception:
Room 9, 7/F, Tung Ning Building, 2 Hillier Street, Sheung Wan, Hong Kong

Factory:
Room 2, 9/F, Sky Star Centre, 35 Hung To Road, Kwun Tong, Kowloon, Hong Kong
Send Email