Hybrid AC/DC Data CentersProject 5: Evaluation of Wide Bandgap Power Semiconductor

Devices for DC-Powered Data Centers

Lead PI: Prof. Alan Mantooth, Prof. Juan Carlos Balda

University Collaborators: Ramchandra Kotecha, Yuzhi Zhang

Potential Industrial Mentors: dcFUSION, EMERGE, EDCS Power,Emerson Network Power, Pika Power, Eltek, Wolfspeed

Project 5 Summary:

Wide Bandgap Device Evaluation

- Perform device evaluation studies of Si, SiC, and GaN for utilization in

various conversion stages

- These can be broadly classified into three voltage categories: (a) Between

1700 Vdc and 600 Vdc, (b) Between 600 Vdc and 48 Vdc, and (c) Below

48 Vdc. Different power semiconductor devices are available for each

category.

Server

AC/DC DC/ACUtility Line

Voltage

480 V3fAC

96% x 97% x 90% x 86% = 71%

PDU120 V1f

AC

PSU

AC/DC DC/DC

12 V

Ele

ctro

nic

Lo

ad

s

VR

VR

Fans

Utility Line

Voltage

480 V3fAC

400 Vdc 400 Vdc

Server

PSU

DC/DC12 V

Ele

ctro

nic

Lo

ad

s

VR

VR

Fans

DC/DC48 V

99% x 99% x 99% x 99% x 94% = 90%

AC-DC Converter

PDU

• AC distribution

architecture, overall

efficiency 71%.

(source

http://deltaww.com)

• 400VDC

distribution

architecture, overall

efficiency 90%.

1) A Compact SiC Power MOSFET model has been developed at U of A

2) The model has been validated against commercial CREE devices rated up to 1200 V

3) A compact GaN HEMT device model has been developed at U of A

4) The model has been validated against commercial EPC devices

5) Several wide band-gap devices have been evaluated for use inside the conversion stages of DC powered datacenter

6) The compact models will be used for design and simulation of intermediate topologies of DC powered datacenter

Accomplishments to date

Accomplishments to date (cont.)

CV model Fits: EPC2021 Commercial GaN HEMT

DC IV model Fits: EPC2021 commercial GaN HEMT

Improve the model-fidelity of SiC and GaN device

compact models

Validate the models for a wide-range of commercial

devices

Evaluate intermediate conversion stages of DC-powered

datacenter through model-based simulations

Develop a vertical GaN device model (contingent on the

availability of devices)

Future Activities

Hybrid AC/DC Data CentersProject 6: Integration of Distributed Energy Sources in DC-

Powered Data Centers

Lead PI: Prof. Juan Carlos Balda, Prof. Alan Mantooth

University Collaborators: Yuzhi Zhang

Potential Industrial Mentors: dcFUSION, EMERGE, EDCS Power,Emerson Network Power, Pika Power, Eltek

Project Objectives To evaluate the integration of suitable technologies of distributed energy sources into

dc- powered data centers.

To identify potential technical and cost barriers.

System Level Configuration 400 V dc-powered data center.

Solar power and hybrid energy storage integrated in the 400 V dc bus.

A cascaded bridgeless PFC multilevel converter with high efficiency, fewer number of devices and high reliability serves as interface to the grid.

Project Summary

400 V dc-powered green data center

AC Bus

400VDC

Grid 2

Grid 1

Diesel Generator

S3

S2

S1

PV Array Wind Farm

Critical Fans and Lighting

Transmission line

Green

Energy

Room AC

SST Structure

S4 Hard drives

Fan

CPU

Memory

48VDC

RACK 1

Battery I1

Hard drives

Fan

CPU

Memory

48VDC

RACK n

Battery In

Battery II

Ultracaps

Solar Panels

(1) The battery and ultracapacitor are combined as hybrid energy storage to improve the stability of dc bus.

(2) A cascaded bridgeless PFC multilevel converter as power interface from grid to data center is proposed.

(3) Model predictive control is implemented to improve the efficiency and transient performance.

(4) The simulation of different percentages of distributed solar power in data centers are performed to analyze the system dynamic response.

(5) One paper is presented in the INTELEC 2016 at Austin, Taxes. Paper title: Ultracapacitor Application and Controller Design in 400 V DC-Powered Green Data Centers.

Accomplishments to date

Considering 3% voltage ripple, in a 0.5 MW data center, 0.25 MW of solar power (50% of data center power) is acceptable for dc bus stability.

In 1.5 MW and 3 MW data centers, the acceptable solar power are about 35% and 42% of data center power rating, respectively.

The tradeoff is the cost of dc bus capacitor and battery.

Different Percentages of Solar Power in Data Center:

Model Predictive Control to improve the stability of 400 VDC bus:

vo1

vo2

iloadvo1

vo2

iload

Conventional PI control Model predictive control

Accomplishments to date (cont.)

Apply model predictive control and design new phase-

shift method in Dual-Active-Bridge to further reduce the

switching loss

Grounding system design and fault detection in 400V dc-

powered data center

Design a scale down prototype with real servers load

Written final report and journal papers submission

Future Activities

Hybrid AC/DC Data CentersProject 7: Solid-State Technologies for Fault Protection in

DC-Powered Data Centers

Lead PI: Dr. Juan Carlos Balda

University Collaborators: Dr. Cheng Deng & Witness Martin

Potential Industrial Mentors: Emerson Network Power, EDCS Power,dc Fusion, Emerge Alliance

Project Objectives Evaluate current solid-state circuit breaker (SSCB) technologies suitable for dc-power

data centers

Hybrid SSCB may not meet power density specification

Propose a SSCB topology suitable for dc-powered data centers

SSCB Specifications Isolating faults within few microseconds

Low voltage drops under normal operating conditions

Effectively handle over-voltages resulting from opening operations

Having a power density of at least 250 W/in3

Project Summary

Topology of a generic dc-powered datacenter

48 VDC Bus

PrintersMonitors

DC

DC

POL Converter

48 VDC Bus

DC

DC

48V

Battery

+

-

Diesel Gen.

Dirty Bus 400 VDCClean Bus 400 VDC

Potential Circuit Breaker Position

Bus

Converter

• • •

48V

400V

48V

GridUPS #1

• • • ComputersDrives

DC

DC

POL Converter

48V

• • • KeyboardsNetwork

HUBs

DC

DC

POL Converter

48V

• • •

Rack #1 Rack #n

• • •

DC

DC

DC

DC

DC

DC

DC

AC

DC

AC

+

-

DC

DC

DC

DC

DC

DC

DC

DC

Fire Sup. HVAC Lights AUX

GridUPS #2

DC

AC

+

-

1V / ••• /12V 1V / ••• /12V 1V / ••• /12V

N/O CB

-200V+200V 0

Diesel Gen.

DC

AC

-200V+200V 0-200V+200V 0 +200V -200V0

• Main goal: evaluate different cases of overvoltage protection for SSCB

Proposed Test Circuit

Item Value Item Value

Vdc (V) 200 R1 (Ω) 47

Lstr2 (μH) 150 C1 (nF) 250

Rload (Ω) 20 D1 STTH5L06RL

S1 IXYH60N90C3 MOV1 S20K175E2

D2 STTH5L06RL MOV2 S20K175E2

MOV3 S20K25AUTO

Proposed test circuit and the prototype

TABLE I Parameters of the prototype

Vdc

S1

Rload

iin

Cin Fault

Lstr2B E

G

FA

SSCB

iclamp

MOV3

D2

Waveforms for overvoltage clamp realized by a series combination of a MOV and a diode on load side of SSCB

Time Scale: 10 μs /div

iin ( 5 A/div) vBE ( 100 V/div)

300 V 15 A

iin ↓ 44%

vBE ↓ 75% Time Scale: 10 μs /div

iclamp ( 5 A/div) vFG ( 100 V/div)

80 V12 A

vFG ↓ 92%

ItemClamp Across SSCB Clamp Across Load Side

Case 1 Case 2 Case 3 Case 4 Case 5 Case 6

Number of

Components3 1 4 1 1 2

Response Time

(μs)25 27 20 8 25 20

Energy Absorbed

(A) or Dissipated

(D)

D D + A D + A D None D

Percent of Vce

Peak Reduction77% 69% 85% 62% 92% 92%

Comparison of Different ClampsTABLE II Indexes comparison of six cases

The series combination of MOV and diode provides better overvoltage protection for the SSCB assembly

Continue with the analysis of the load-side results

Incorporate model of a rack as the load

Analyze effects of source-side inductance

Propose a preferred SSCB configuration

Future Activities

Hybrid AC/DC Data CentersProject 8: Fast Arc Detection in DC-Powered Data Centers

Lead PI: Dr. Juan Carlos Balda

University Collaborators: Dr. Cheng Deng & Dimas B. Fiddiansyah

Potential Industrial Mentors: Emerson Network Power, dcFusion,Emerge Alliance

The main objective is developing a simple, accurate, reliable and

cost-effective method for dc arc detection that could be integrated

into a SSCB controller.

Project Objective

DC Arc Circuit Model

Vdc

vq

Rgap

egapigap

RLoad

vgap

Fig. 1 (a) Arc circuit model, and (b) Matlab/SimulinkTM results of gap voltage and current trajectories.

(a) (b)

gap q gapv v e

Iarc

Varc

Two Possible of the Arc Detection Methods

• Transforms time-based signals to

frequency-based signals

• Provides information about frequency

components but not time occurrence

of an event

• Space and frequency analysis

(scale and time)

• Converts a signal into a series of

sub band signals

2. Discrete Wavelet Transform (DWT)

1. Discrete Fourier Transform (DFT)

Wavelet output using the Daubechies order 4 (db4) that was decomposed from the arc fault current signal

Optimizing the design of the proposed the DC arc model using Matlab/Simulink™

Refine the dc arc detection system algorithm

Develop a laboratory prototype

Future Activities

Thank You for Your Attention!