Texas Devices Inc. (TI) introduced a number of energy administration gadgets and a reference design to assist corporations meet AI computing calls for and scale energy administration architectures from 12 V to 48 V to 800 VDC. These merchandise embrace a dual-phase sensible energy stage, a dual-phase sensible energy module for lateral energy supply, a gallium nitride (GaN) intermediate bus converter (IBC), and a 30-kW AI server energy provide unit reference design.
“Knowledge facilities are very advanced techniques they usually’re operating very power-intensive workloads that demand an ideal stability of a number of crucial elements,” stated Chris Suchoski, common supervisor of TI’s information heart techniques engineering and advertising and marketing workforce. “Most vital are energy density, efficiency, security, grid-to-gate effectivity, reliability, and robustness. These elements are notably important in creating next-generation, AI purpose-driven information facilities, that are extra power-hungry and important at present than ever earlier than.”
Suchoski describes grid-to-gate as the entire energy path from the AC utility gird to the processor gates within the AI compute servers. “All through this path, it’s crucial to maximise your effectivity and energy density. We will help enhance total vitality effectivity from the unique energy supply to the computational workload,” he stated.
TI is targeted on serving to prospects enhance effectivity, density, and safety at each stage within the energy information heart by combining semiconductor innovation with system-level energy infrastructure, permitting them to attain excessive effectivity and excessive density, Suchoski stated.
Energy density and effectivity enhancements
TI’s energy conversion merchandise for information facilities tackle the necessity for elevated energy density and effectivity throughout the complete 48-V energy structure for AI information facilities. These embrace enter energy safety, 48-V DC/DC conversion, and high-current DC/DC conversion for the AI processor core and facet rails. TI’s latest energy administration gadgets goal these next-generation AI infrastructures.
One of many tendencies available in the market is a transfer from single-phase to dual-phase energy phases that allow larger present density for the multi-phase buck voltage regulators that energy these AI processors, stated Pradeep Shenoy, technologist for TI’s information heart techniques engineering and advertising and marketing workforce.
The twin-phase energy stage has very high-current capabilities, 200-A peak, Shenoy stated, and it’s in a really small, 5 × 5-mm package deal that is available in a thermally enhanced package deal with top-side cooling, enabling a really environment friendly and dependable provide in a small space.
The CSD965203B dual-phase energy stage claims the best peak energy density energy stage available on the market, with 100 A of peak present per part, combining two energy phases in a 5 × 5-mm quad-flat no-lead package deal. With this gadget, designers can enhance part rely and energy supply throughout a small printed-circuit-board space, bettering effectivity and efficiency.
One other associated pattern is the transfer to dual-phase energy modules, Shenoy stated. “These energy modules mix the facility phases with the inductors, all in a compact type issue.”
The twin-phase energy module co-packages the facility phases with different elements on the underside and the inductor on the highest, and it gives each trans-inductor voltage regulator (TLVR) and non-TLVR choices, he added. “They assist enhance the general energy density and present density of the answer with over a 2× discount in measurement in contrast with discrete options.”
The CSDM65295 dual-phase energy module delivers as much as 180 A of peak output present in a 9 × 10 × 5-mm package deal. The module integrates two energy phases and two inductors with TLVR choices whereas sustaining excessive effectivity and dependable operation.
The GaN-based IBC achieves over 1.5 kW of output energy with over 97.5% peak effectivity, and it additionally allows regulated output and lively present sharing, Shenoy stated. “That is vital as a result of as we see the facility consumption and energy masses are growing in these information facilities, we’d like to have the ability to parallel extra of those IBCs, and so the present sharing helps make that very scalable and simple to make use of.”
The LMM104RM0 GaN converter module gives over 97.5% input-to-output energy conversion effectivity and excessive light-load effectivity to allow lively present sharing between a number of modules. It could actually ship as much as 1.6 kW of output energy in a quarter-brick (58.4 × 36.8-mm) type issue.
TI additionally launched a 39-kW dual-stage energy provide reference design for AI servers that incorporates a three-phase, three-level flying capacitor power-factor-correction converter paired with twin delta-delta three-phase inductor-inductor-capacitor converters. The facility provide is configurable as a single 800-V output or separate output provides.
TI additionally introduced a white paper, “Energy supply trade-offs when getting ready for the subsequent wave of AI computing development,” and its collaboration with Nvidia to develop energy administration gadgets to assist 800-VDC energy architectures.
The options can be on show at Open Compute Summit (OCP), Oct. 13–16, in San Jose, California. TI is exhibiting at Sales space #C17. The corporate will even take part in know-how classes, together with the OCP International Summit Breakout Session and OCP Future Applied sciences Symposium.
The submit TI launches energy administration gadgets for AI computing appeared first on EDN.
