PREDICTION, AN UNCERTAINTY - Electrotechnische ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Electrotechnische Vereeniging Issue 24.2 | March2021
PREDICTION, AN UNCERTAIN T Y
Interviewwith Dr.ir. G.R. Chandra Mouli
About EV Charging and load prediction
Black Hole Visualization
By Annemieke Verbraeck
What is Quantum Computing?
By Prof. dr. ing. K.L.M. Bertels
IN MEMORIAM
We deplore to announce that has passed away at an age of 96
Prof. ir. Hans van Nauta Lemke
Emeritus Professor and Rector Magnificus Delft University of Technology
Honorary Member of the Electrotechnische Vereeniging
* †
November 22, 1924 January 10, 2021
Palembang ‘s-Gravenhage
Since 1989 the Electrotechnische Vereeniging has had the honour of having Mr.
Van Nauta Lemke as Honorary Member. Our thoughts go to all involved.
24.2
From the Board
Dear Reader, Dear Reader,
Another Maxwell edition is pub- students who are actively enjoying We have arrived at a new quarter.
lished and thus another quarter has the ETV during these times. I don’t think anyone would have
passed in this very strange time. It This year has taught me and the been able to predict the current
has been tough for all of us. We are rest of the Board to make the best situation at the start of this aca-
all familiar with the circumstanc- out of a situation that did not go as demic year. EWI has been closed
es, and we all would like to have a anticipated because as we all know, for a few months now, and prob-
taste of our old lives again. Just for this year played out differently than ably will be for a bit longer due to
one day, to break free from the daily other years. fire safety issues. Luckily, there
grind we have lived in the previous is light at the end of the tunnel,
year to shake things up. A few weeks ago, the EEMCS Re- as experts expect the virus to be
cruitment Days have taken place, containable by this summer. This
Some of us have discovered new which to me was the most impor- means that activities such as the
hobbies, some of us have found tant event this year. It is something first year weekend, the gala and
a new love for being outside and I was extremely excited about and Lustrum activities are just over
some have enjoyed watching lots I have worked towards since this the horizon. Keep your head up!
of movies with their housemates. year started. If you have any questions or com-
The pandemic has made us more Now this intense yet special week ments regarding your studies,
creative and forced us to discover has passed, I have more varied du- know that we work daily from our
new things we did not know about ties which means there is room to digital Board room on discord. You
ourselves. work on new projects and coming are always welcome to join us for
up with fresh ideas to introduce our a chat even if it is not study relat-
A year ago I was asked if I were partners to the students. ed!
up to the challenge of becoming a
Board member of the Electrotech- Enjoy the rest of the year, and re- With excited regards,
nische Vereeniging, and what an mind yourself to relax once in a Tim Plantfeber
experience has it been so far! while. Secretary of the 149th Board of the
Electrotechnische Vereeniging.
When looking back on the past few With excited regards,
months I see a lot of exciting activ- Floor Walterbos
ities, a lot of new committees and Commissioner of External Affairs
lots of of the 149th Board of the Electro-
technische Vereeniging
March 2021 3
24.2
Colophon Editorial
Dear readers,
Year 24, edition 2, March 2021
Welcome to the second edition of the 24th Maxwell! The theme of this
Editors edition revolves around the role of electrical engineering in not only
Yunus Emre Döngel building, but also predicting, our tomorrow.
Merel Verhoef
Aniruddh Kulkarni The power of prediction in the mathematical sciences, stemming from
Natasha Birari the understanding of probability together with access to massive
Nitish Kulkarni amounts of data, has become a cornerstone of modern scientific ap-
plications. Unfortunately, in the public sphere, it has also given birth to
Designers one of the biggest modern misconceptions about science: that scientif-
Neil Dani ic outcomes are ‘conclusive’, whereas in fact, all scientific methods are
essentially the understanding and art of ‘uncertainty’.
Contributors
K.L.M. Bertels, G.R. Chandra Mouli, For this edition, the Maxwell committee reached out to researchers
Kees Pronk, Annemieke Verbraeck, who are working with prediction and uncertainty across diverse fields
Simon Tindemans, Jan de Jong, in electrical engineering – from black hole visualisation to quantum
Gijs Lagerweij, Tim Plantfeber, acceleration, to how data-based prediction is being incorporated into
Floor Walterbos EV charging and the European power grid.v
Advertisement We hope you will enjoy the articles we have brought together. A big
page 32 Witteveen+Bos thank you to all our contributors for sharing their experience and in-
sights. I also want to express my gratitude to the Board and my fellow
Printing committee members, without whom this edition would not have been
Quantes, Rijswijk, 1400 copies possible.
Electrotechnische Vereeniging Natasha Birari
Mekelweg 4
2628 CD Delft
The Netherlands
Phone number: +31 15 278 6189
Email: board-etv@tudelft.nl
Website: etv.tudelft.nl
149th Board of the ETV
President - Joris van Breukelen
Secretary - Tim Plantfeber
Treasurer - Vera Pauptit
External Affairs - Floor Walterbos
Education - Reinier van der Leer
Subscriptions
Non-members can get 4 issues of the
Maxwell against an annual contribution of
€10,-.
Change of address
Please send changes in your address to
the (e-mail) address above or use the
website.
4 March 2021
24.2
INDEX
6 The Hammer
Computer
10 EV Load Prediction
14 Black Hole
Visualization
20 A physicist who never
lost her humanity
22 Student Columns
26 What is Quantum
computing?
30 Possibly the D-Day
March 2021 5
24.2
The Hammer Computer
Kees Pronk
At the Study Collection, we were structure of this analogue and
aware of a large machine (3.0 x mechanical calculator.
1.5 x 1.0 meters) standing in a dark
corner. The origin and purpose of The construction of this machine
this machine had long been for- started around 1950. To really
gotten. Fortunately, an observant know what we are dealing with,
visitor was able to tell us where we must go back to that time to
the device was built and who the understand the development of
designer was. radar technology and the con-
The device had been constructed struction of computers in those
around 1950 at the Nederlands years. This whole exercise re-
Radar Proefstation (Dutch Ra- sembles technological archaeol-
dar Testing Station) in Noord- ogy: from the artifacts found, an
wijk (now: Christiaan Huygens image is formed of the operation
Laboratory in Katwijk) under the of the machine, like a puzzle be-
supervision of its designer ir. J. ing slowly put together.
A. Hammer (figure 1). A brief de-
scription of the machine, as well The Hammer computer was de-
as the author, is given in [3]. signed and built for a special pur-
A literature study yielded two pose: the design and calculation Fig 1. Hammer’s
articles. The first article [1] is a of linear radar antennas. With Calculator
report by Hammer at a scientific radar transmission antennas, it
conference in France, expound- is important to generate a beam
ing the scientific progress made of radiation that is as narrow as tributions from all slots. Due to
by using this machine. In the ac- possible, without so-called “side interference, the fields generat-
companying survey article [2] lobes” (figure 2). Since the radi- ed by the individual slots amplify
more details of the machine are ation pattern of a single antenna each other in the antenna aiming
presented. After studying these is relatively wide, an array of an- direction and weaken each other
articles and opening up the ma- tenna-slots is used to achieve a in the other directions, thereby
chine to examine its construc- narrow beam (figure 3). The total reducing the effect of the side
tion, gradually, a picture began field generated by these anten- lobes.
to emerge of the purpose and na-slots is the sum of the con-
Antenna aiming direction
Main Lobe
Side Lobe Side Lobe
Side Lobe Side Lobe
Fig 2. Radiation Diagram of a single Fig 3. Example of a slotted antenna
antenna
6 March 2021
24.2
The Design of the Computer slots. In other words, this unique a signal distribution device These
Hammer’s computer is designed mechanical / analogue machine channels are numbered -20° .. 0°
to perform two functions: (i) using simulates radar waves. The .. +20°. Each channel simulates
up to 41 antenna slots, the machine block diagram of the computer one antenna slot.
calculates the field strength can be seen in figure 4. Whereas
of the E-field, and (ii) starting the frequency range of radar is
from a prescribed radiation around 30 GHz, here a frequency
pattern, the machine calculates of 1000 Hz is used. The sine wave
the needed contributions of the of 1000 Hz is presented to 41
(at most 41) individual antenna parallel calculation channels via
Fig 4. The block diagram of Hammer’s machine. All calculations assume that the antenna slots
are the same distance ‘d’ from each other
Fig 5. X-Y resolvers and the gear box. The positive resolvers rotate clockwise; the negative
counter clockwise. The central resolver is stationary
March 2021 7
24.2
To calculate the E-field the oper- gearbox is driven by a central cessing. Towards this purpose, a
ator manually sets the amplitude motor. This motor sets the angle so-called ‘rootpot’ is connected
and phase of each antenna slot θ from the aiming direction. to the circuit [2]. The functioning
on the computer. The amplitude Using the formula of this part is based on finding
for each channel can be set with the roots of the equation
a buffered 10-turn helipot.
Some more thought was put into
setting the phase on this ma- the strength of the E-field of the
chine. Each channel is equipped nth antenna slot is calculated. Solving this equation using the
with a so-called x-y resolver. The Subsequently, all field strengths ‘rootpot’ will give a number of
experimenter sets the desired are added together: complex roots. Although this part
value of the phase by manually has been located in the machin-
turning the stator of the resolv- ery (figure 7), its precise opera-
er. The rotor of each resolver is tion is still unclear. Presumably
driven from a central gearbox The result is shown to the exper- the ‘rootpot’ is also used as the
with ratios 1: 2: 3: ..: 20. When the imenter via an oscilloscope and/ aforementioned signal distribu-
desired phase is set, the x-y re- or a paper recording device (re- tion device.
solver translates the mechani- member when paper recording
cal angle between the stator and was a norm?) as shown in fig- Hammer’s computer calculates
the rotor into two AC voltages ure 1. The addition of the signals the E-field in 12 seconds. Digital
[5]. Both, the resolvers and the is done via buffered amplifiers computers of that time (in 1955)
gearbox, have been mounted on that use vacuum tubes. Figure needed 700 seconds for this cal-
a very sturdy sub-chassis, so 6 shows the bottom part of the culation. The great developments
as to increase the accuracy of electronic signal processing unit that took place in digital tech-
the computation. Figure 5 shows of eight resolvers. nology later made this analogue
some resolvers and the gearbox. / mechanical calculator from
The computer’s second function Hammer superfluous, and it was
The E- field is calculated for a of is calculating the amplitude eventually donated to the Study
large number of angles and phase settings of the anten- Collection in Delft.
(θ = -30° , ... , + 30°) from the na slots starting with the pre-
antenna aiming direction). The scribed E-field i.e. inverse pro-
Fig 6. Electronics for signal Fig 7. The Rootpot
processing of 8 resolvers
8 March 2021
24.2
Applications of Linear Radar used here. For more information ed scientific and applied work on
Antennas on modern techniques, the read- this subject. One of the develop-
Various radar installations were er is referred to [4]. ments was the creation of this
constructed on the basis of ex- analogue / mechanical computer
periments with Hammer’s com- Conclusion by ir. J. A. Hammer in the years
puter, both onshore and on frig- During the Second World War, the 1950-1955. It is remarkable that
ates of the Royal Dutch Navy. development of radar technology Dutch engineers were able to
Figure 8 shows an application in the Netherlands came almost create this splendid example of
of a linear radar antenna on the to a standstill. After the war electronic and mechanical design
Dutch frigate zr. ms. Tromp (the was done, it was decided that so shortly after the end of WWII.
black box on the stern, designed the Netherlands should start its
and produced by Thales in Hen- own developments in this field.
gelo). Of course, the most mod- The Dutch Radar Test Station was
ern radar techniques have been established in 1947, and it start-
Fig 8. zr. ms Tromp
Literature:
[1]The Evaluation of Diffraction Prob- thesis of Linear Arrays with Prescribed [4]https://www.museumwaalsdorp.nl/nl/
lems in Aerial Technology by Means of an Radiation radar-nl/radar-phased-array-onderzo-
Analyser for Patterns: A Survey of some Early Work ek/
Complex and Fourier Series, by: J. A. at the Christiaan Huygens Laboratory, by:
Hammer in: Principes fondamentaux: E. Goldbohm (formerly professor at [5]X-Yresolver: https://en.wikipedia.org/
Équations de Maxwell, Principe de Huy- TUDelft) in: IEEE Antennas and Propaga- wiki/Resolver_(electrical)
gens et Théorie de la diffraction en hyper- tion Magazine, Vol. 35 No. 3, June 1993 tennas and Propagation Magazine, Vol. 35
fréquences (1956) No. 3, June 1993
[ 3 ] h t t p s : / / w w w.c r y p t o m u s eu m .c o m /
[2]Near Field Measurements and the Syn- manuf/nrp/files/nrp_40.pdf
March 2021 9
24.2
EV Charging: Load Prediction,
Smart Charging and its Impact
on the Electric Grid
Interview with Dr.ir. G.R. Chandra Mouli
Aniruddh Kulkarni
Dr. G.R. Chandra Mouli is an As- face of the automobile market:
sistant Professor in the DC sys- the electric vehicles. In 2010, the
tems, Energy conversion and global count of electric cars was
Storage group in the Department only 17,000. As per the Interna-
of Electrical Sustainable Energy. tional Energy Agency, the auto-
He has received his PhD from mobile market has seen a year-
the Delft University in 2018 on on-year growth of 6%, swelling
the dissertation associated to the total count of electric cars to
the development of a solar pow- 7.2 million1.
ered V2G electric vehicle charg-
er compatible with CHAdeMO, This exponential rise in the num-
CCS/COMBO and designed smart ber of electric vehicles, raises an
charging algorithms. Dr. G.R. important question: Is the charg-
Chandra Mouli received ‘Under 35 ing infrastructure developing at
Innovator’ award from the IEEE par with the vehicle technology?
Industrial Electronic Society in To this Dr. G.R. Chandra Mou-
2019. Based on the specialization li puts forth a perspective that
of Dr. G.R. Chandra Mouli, we dis- this development of charging in- ance is maintained in the ratio of
cuss the future of the electric ve- frastructure is dependent on the number of vehicles to the charg-
hicle charging infrastructure. The country. Development of charging ing stations. This development in
21st century has seen the new infrastructure means that, a bal- the charging infrastructure var-
Fig-1: Rise of EV market from 2010 to 2019
10 March 202124.2
ies by geography. According to in installing and operating charg- tering to the peak demand of
the International Energy Agency, ing stations are Fastned, Ionity, the load. The peak demand cy-
the global figure for installed Tesla Supercharger, EVgo. cle occurs at fixed durations
chargers were 7.3 million in 2021. in a 24-hour cycle. Congestion
There are a few countries which The load experienced due to Management deals with spread-
have more vehicles than the charging EV’s has much higher ing out the time during the peak
charging infrastructure. Figure-2 magnitude than the conventional load hours throughout the 24-
shows country-wise distribution loads. Hence, the impact on the hour period such that the energy
of number of charging stations electrical grid will be much high- consumed remains the same but
built. This imbalance results in er when the EV-charging is taken the power reduces for the peak
Publicly accessible slow chargers 598,000 Publicly accessible fast chargers 264,000
Fig-2: Global distribution of publicly available charging stations Publicly accessible fast chargers
264,000
an increased waiting time for us- into account. Most of the pres- load time intervals. An alter-
ers to charge their vehicles or to ent distribution electrical grids nate method is to install locally
commute to large distances due are not capable of handling such generated renewable charging
to unavailability of nearby charg- kinds of loads. For reducing this stations. This will be effective in
ing stations. impact on the grid techniques like sense that there will be less im-
“Smart Charging” have come to pact of the load on the grid.
The charging infrastructure can existence. Smart charging aims Aggregator based Charging: -
be categorized into two types: at managing the supply and de- One of techniques of smart
The first type of charging sta- mand of the EV charging, by es- charging is the aggregator-based
tion is a long duration charging tablishing protocols/algorithms charging. Aggregator based
station i.e., chargers located in to the way the users charge their charging is one of the techniques
offices, public charging stations vehicle and control the way in which involves the contribution
within the city etc. Such sta- which battery of an electric ve- of the whole EV ecosystem. The
tions are meant for fulfilling the hicle is charged/discharged. This EV (electric vehicle) ecosystem
daily charging requirements of is one of the focus points of the consists of several key compo-
EV owners. The second type of Department of Electrical and nents: the car manufacturers,
charging stations are meant for Sustainable Energy. One of the the charger manufacturing in-
fast charging. These stations are techniques that Dr. G.R. Chan- dustries, industries that install
located on highways and routes dra Mouli confers is Congestion and operate the chargers, indi-
of long-distance travel. Some of Management. This in other words vidual entities that provide the
the industries making progress is the efficient management ca- charge controlling software. The
March 2021 1124.2
energy market players such as demand of every individual EV not included in the aggregator
energy producers and suppliers, user. This charging information supervision. The load at such
companies responsible for sup- is dependent on the user profile charging stations can be ac-
ply-demand balance and the dis- by collecting information such counted for as uncertain load.
tribution system operators also a as arrival time, parking time. Out Hence, the net load predic-
part of the EV ecosystem. of the whole EV ecosystem, any tion is made by observing the
An individual EV owner/user does entity can take up the role of an charging pattern of EV owners
Traditional wide beam communication systems vs next generation system with
dedicated for users
not have any idea of the total load aggregator. Independent entities and extrapolating the data.
experienced by the grid at certain can also assume the role of an The load due to EV charging is
point of time. Hence, from a cus- aggregator. Multiple aggregator a long duration load imposed
tomer centric view the control of companies are present in the on the grid as compared to the
charging is not possible. Based ecosystem which collect the data conventional loads, which are
on the user behaviour, of a set of users for predicting switched frequently. Due to the
there are certain durations in a and controlling a part of the load. prolonged time for charging,
day in which most of the EV’s of Collectively, this data presents there will be an overlapping in
a particular region will be charg- the total load profile at a national the loading conditions i.e., due
ing. This will result into a surge grid level. to multiple users charging their
of the load on the grid. These in- vehicles in the same time dura-
termittent surges of load during To acquire the data of individual tion. Once the load profile is built
the peak demand duration have charging demand, the aggrega- based on integrating individual
to be distributed in order to re- tors use different charging pro- charging demand, in the same
duce adverse impact on the grid. tocols. Some of the charging pro- time duration. Based on the grid
To accomplish this distribution tocols used by the aggregators congestion and available sourc-
there will be some potential en- are Open Charge Point protocol, es such as renewable charging,
tities in the future, who will be the Open Smart Charging proto- different pricing schemes can be
responsible to cluster the data col. For example, the e-mobility devised. The cost for charging is
and provide a prediction of load, service provider handles the fi- determined based on the optimal
spread out in a 24-hour cycle. nancial transactions related to power consumed versus time
These entities are called “Aggre- charging. This financial trans- profile for every individual vehi-
gators.” action reflects the details com- cle. Hence, one of the possible
prising of duration of charging, methods is to decrease the load-
The primary function of an aggre- location of charging etc. Howev- ing during peak demand is to in-
gator is to acquire the charging er, some charging stations are crease the price charged per unit
12 March 202124.2
power consumed (cost/kWh). mand [2]. This data concerns with installed capacity.
the grid loading conditions. For an The above situation describes
Despite the smart charging tech- individual, the maximum demand that the grid will certainly need
niques, there are multiple factors will increase by 18% from every an upgrade in the power capacity
determining the loading on the 10% increase in houses with EV’s to satisfy the growing demand of
grid such as EV penetration, grid [3]. The impact of this increased EV charging. However, upgrading
features, rated charging power of load will be more prominent on the grid will require a significant
vehicles, charging behaviour pat- the distribution network than the investment. Hence, Dr. G.R. Chan-
tern etc. Based on the EV pene- transmission network. Most of dra Mouli believes that the future
tration the peak load for grid has the distribution network have an of charging infrastructure will
been calculated. Research has installed capacity to operate at turn towards an optimal solu-
shown that a 10% market pene- higher load conditions, there are tion which combines the smart
tration of EV would result in an some networks present which charging techniques and the up-
increase of 17.9% in daily peak operate near to the peak load. grade in electrical gird.
demand while a 20% market pen- Such networks, will undergo
etration would result in an in- overloading condition. This issue
crease of 35.8% in daily peak de- can be resolved by increasing the
References: -
1. IEA (2020), Global EV Outlook 2020, IEA, vol. 26, no. 2, pp. 802-810, May 2011, doi: Dearborn, MI, USA, 2009, pp. 827-831, doi:
Paris https://www.iea.org/reports/global- 10.1109/TPWRS.2010.2057456. 10.1109/VPPC.2009.5289760.
ev-outlook-2020
3. G. A. Putrus, P. Suwanapingkarl, D.
2. K. Qian, C. Zhou, M. Allan and Y. Yuan, Johnston, E. C. Bentley and M. Narayana,
“Modeling of Load Demand Due to EV “Impact of electric vehicles on power
Battery Charging in Distribution Systems,” distribution networks,” 2009 IEEE Vehicle
in IEEE Transactions on Power Systems, Power and Propulsion Conference,
March 2021 1324.2
Black Hole Visualization
Annemieke Verbraeck
A surprising amount of black-
hole articles
It seems every week I find a new
article on my feed concerning
new and surprising discover-
ies in space. Many of these are
about black holes, although that
might be just me, as Google must
think I am obsessed with the
phenomenon after searching for
the Black Hole Wikipedia article
hundreds of times during my re-
search. Even though black holes
are surprisingly simple - some
of them can be defined with only
one parameter – their influence
on their surroundings is not easy
to grasp. Despite my background
in Astronomy, I often found my-
self double checking the basics
to make sure my implementa-
tions and definitions were cor-
rect. If my Astronomy BSc taught
me anything, it is how easy it is
to make mistakes dealing with Copyright© Frank Auperlé
complicated physics formulas
that cannot be verified in an ex- visualization, and concluded that
periment on earth. their simulation was not optimal, Taking a trip to a black hole
at least in terms of speed. With The easiest is to start the expla-
From Astronomy to Computer my astronomy background, I was nation of our research with a little
Graphics the perfect candidate to fully un- thought experiment to describe
Another thing the Astronomy BSc derstand all the formulas in the our setting. Imagine yourself to
taught me, is that those physics paper and work out a smarter be a wealthy tourist on one of the
formulas were a bit too abstract and faster solutiovn. I am now in first spaceship trips. The guide
for me and a lot more present in my third year of my PhD at the starts talking excitedly, as you
the study than the pretty Hubble Computer Graphics group, and are approaching the grand fina-
Telescope imaging of deep space. only this October did we publish le of the tour, a real Kerr black
This realization led me to pursue a a paper on our findings. As it of- hole. Unlike Schwarzschild black
master in computer science with ten goes in research, at first it is holes, Kerr black holes are spin-
a specialization in Graphics at more difficult than expected and ning rapidly, so the spaceship
the TU Delft. For my masters pro- at the end there is always some has entered a stable orbit around
ject, the professor of our group, code to improve and some func- it, to prevent getting sucked in.
Elmar Eisemann, immediately tionality to add. The end result After all this travelling you know
knew what I was going to spent was worth all the work howev- that space is a very empty place,
all my time on. He saw the mov- er as we received an honorable so while you are close to the black
ie Interstellar, skimmed through mention for best SciVis paper at hole, all the stars and galaxies
the paper about their black hole the Vis 2020 conference. are almost infinitely far away. So
14 March 202124.2
far away in fact that if you would fore we can go on touristic trips curved paths through space. As
make a simulation, they could to a black hole, so until then we we only detect the photons when
be approximated by existing on have to be satisfied with simula- they reach our eyes, we expect
a celestial sphere with infinite tions, but not everyone has a CPU the star to be located in the di-
radius around you and the black farm at their disposal. Our ap- rection where the photon arrived
hole. The big question now is of proach therefore aims to produce from. On top of that you can see
course, what do you see when a high-quality interactive simula- that different light rays from one
you look out the window? What tion that can be run on your own star follow different paths to your
do you expect to see? computer in real time, to come a eye, which means photons from
little closer to that experience of the ame star come in at different
Real image versus Simulation looking out of the window of the angles, and thus show you im-
The first thing that now pops spaceship. ages of the same star at differ-
into your mind is probably the ent locations in Fig.2. To make it
image of our simulation (Fig.1) I will explain step by step how even more crazy, the multiplying
that made you interested in this we achieve this, but before that, effect happens an infinite num-
article in the first place. Had you we first must understand why ber of times. And then some light
not seen it however, it might be visualizing a black hole is not rays do not reach your eye at all,
the image of an existing black straightforward and why it took as they are trapped by the black
hole that was constructed last Interstellar so long to render hole and cannot escape. This re-
Fig 1: Result of the simulation
year from telescopic data, or it these frames. sults in the black portion in im-
might be a simulation, like the agery of black holes, called the
one of Gargantua, the black hole Infinite Images shadow. Looking at it from this
in Interstellar. A large team of As mentioned before, black holes way, saying that we aim to visu-
researchers worked on its de- influence their surroundings. The alize a black hole is not accurate
piction - one of which is the fa- presence of a massive body like at all, as the only thing we see –
mous astrophysicist Kip Thorne a black hole deforms space-time, or not see – is how it influences
- to make sure it was a very and this affects everything that space around it.
precise and physically accurate gets close to it, even weightless
simulation. While this research particles like photons. Let us Light always takes the shortest
produced IMAX quality imagery, imagine following the light rays path
a CPU farm was required for ren- originating from a far-away star Making a virtual world and de-
dering, and for some frames this until they hit your eye. As you can picting it as accurately as possi-
process took up to 100 hours. see in Fig.3, due to the presence ble on a computer screen is what
of the black hole, the rays do not computer graphics research is
Obviously it will take a while be- follow straight lines, but take for. The most basic approach
March 2021 1524.2 uses the physical principle that interactive simulation is not fea- and brightness. The elliptical we can observe an object be- sible when having to trace lots of ray bundle equations are even cause there is light that hits the these rays. This means there are more complicated than single object, and then bounces from two things we had to do: making rays however, and thus take the object into our eyes, enabling the ray tracing faster, and doing even longer to compute. us to construct an image of it in less of it. our mind. We invert this princi- Step 1: Faster Raytracing ple to get images from a virtual Our first step to achieve Fig 2: Due to the presence of the black hole, we see Fig 3: Light rays will always follow the shortest path multiple versions of the same star at different locations. from A to B. world onto a screen. To find out Stellar Problem speedup is to use the principle the correct color for each pixel The researchers depicting the behind the complicated ellipti- we trace a light ray back from the stellar sky behind Gargantua cal bundles, but do the actual viewer, through the pixel, into the quickly found out that their initial tracing using single rays. We virtual scene, to see what object method of tracing one or multi- also work with an area, but it hits and if the object is illumi- ple rays per pixel resulted in a one that starts as a square, the nated. However, these scenes completely black image. The rea- pixel itself. We then trace a ray are usually in an earth-like en- son for this is that the stars are through every pixel corner and vironment, where light rays fol- defined as point lights and hitting end up with a polygonal region low straight lines, and we can in this exact point with a traced ray on the celestial sky, which we general expect objects to be lo- is almost impossible. While larg- can integrate just like the el- cated at the direction we observe er objects like galaxies can be liptical area. As every corner them in. For our black hole case, projected onto a 360-panorama is shared by 4 pixels, we get we do take a similar approach, map that covers the celestial sky to trace an area at the cost of where we start at the viewer and (Fig.5), stars are much smaller only one ray per pixel. With our trace rays through the pixel back lights than even one pixel. The polygonal regions we trade in a to the Celestial Sky to see what solution for their missing stars bit of quality for a lot of speed- star or galaxy it hits there. Due was not to trace single rays, but up. to the presence of the black hole elliptical ray bundles. Every bun- however, this ray will be curved, dle starts from a spherical area In practice the integration and can only be traced through around a pixel, gets traced back at the celestial sky is noth- space by performing a numerical through space and ends up as ing more than a check to see integration of a set of differential an elliptical area on the celes- which stars and which part of equations based on the space- tial sky (Fig.4). All stars in this the 360 panorama is included time metric. This ray-tracing in- area could then be integrated to in the polygon, and then adding tegration is very costly, and an determine the final pixel color up their colors and brightness. 16 March 2021
24.2
Step 2: Less Raytracing the edge of the black-hole shad- need to find the mapping from
For the second step of our speed- ow. A lot of rays are needed in every pixel corner to its loca-
up, we use the fact that the dis- those regions to correctly inter- tion on the celestial sky. With
tortion of the environment stays polate, but in less distorted re- an adaptive grid, we have the
the same as long as we stay in the gions, we can probably get away mapping for each grid vertex to
same place. For one such posi- with much less. To decrease the the celestial-sky location. Sub-
tion, we only compute rays once, number of rays even further, we sequently, we have to locate in
coming in from directions defined construct an adaptive grid that which grid cell a pixel corner is
on a spherical grid all around us. refines more in areas with more located, and then interpolate the
Because the distortion is smooth, distortion. To construct the grid, surrounding values of the grid
we can interpolate this generat- we start with large grid cells and to find the pixel corner’s celes-
ed grid to create images looking use the amount of distortion to tial-sky location. To better ap-
in any orientation and with differ- check if they need to be divided. proximate the smoothness of the
ent projections. This way, we can The end result can be seen in distortion and avoid artefacts, we
Fig4: Top: Elliptical ray bundles by Interstellar,
Bottom: Our polygonal ray bundles made out of single rays.
Fig 5: Left: Celestial sky map, Middle: Map as a panorama image,
Right: Distortion of celestial sky by a black hole, seen in panorama view
show what our eyes would see Fig.6. Depending on the position use Hermite Splines for our in-
with their limited field of view, or of the viewer, the adaptive grid terpolation method.
show how everything around us traces around 40 times less rays
deforms at once, with a 360 pan- than the standard one. What about our trip around the
orama like in Fig.1. black hole?
As you can see in Fig.5, the dis- Interpolation Now we can compute grids, but
tortion is more severe closer to When constructing an image, we unfortunately even with all im-
March 2021 1724.2
provements it takes about half a into too much detail here, our Looking back in time
second for all rays to be traced, solution computes the interpo- When we observe space, we are
so too long for interactive fram- lated shadow and warps the two looking back in time, as it takes
erates. Luckily when we have the existing grids around it – mak- quite a while for light from distant
grids, we can render the images ing it possible to interpolate the stars and galaxies to reach us. As
Fig 6: Adaptive grid in panorama view. Every intersection on this grid represents a light ray that has been traced back to
the celestial sky.
in real-time, by performing this grids at the same positions. I am very busy with a new project,
step in parallel on the GPU. Our the black-hole simulation already
mission was however, to be able And of course, a lot more starts feeling like a thing from the
to feel like we are in a spaceship, This is not all functionality our past. It would be great however, to
travelling around the black hole. method has, but it does outline bring this software to an educa-
If for every position our space- the largest steps that enabled us tional space. I would have enjoyed
ship is at, we have to wait half a to make a high-quality interactive some of my astronomy lectures
second for a new grid, our imple- simulation, while tracing only a a lot more if the professors had
mentation would not be very in- fraction of rays compared to pre- shown me simulations like this!
teractive. To solve this, we come vious solutions. We also made
to our last attempt at simplifi- sure to emphasize the aesthet- Podcast episode with Annemieke Verbraeck
cation and decreasing compute ics of the resulting imagery, we In The ProfCast, Marieke (PhD candidate
time; we aim to compute less are from the Computer Graphics Applied Mathematics) and Dave (communi-
grids. group after all! For stars we use a cations advisor) talk to scientists who want
Gaussian filter to show their dif- to improve the world, who prefer to educate
Step 2.5: Less grids ference in brightness and added young people and who avre looking for that
To simulate movement, without diffraction effects (Fig.8). While one spark. What drives them? Where does
having to recompute the grid the particular pattern we use is their fascination with science come from?
all the time, we came up with a an artifact caused by the vanes And how do they dealv with all the hus-
method to interpolate between of a telescope, we are so used tle and bustle? In recent weeks they spoke
grids at different positions. This to see imagery of space like this, with Annemieke Verbraeck, among others.
is not as simple as it seems, if we that it was a natural addition. We Black holes are one of the most remark-
look at the deformations between also include color and brightness able phenomena in the universe. But what
two grids, like in Fig.7, it is clear changes induced by the black do they look like? And how do you represent
that one produces a larger shad- hole, as it influences the wave- such a thing? Researcher Annemieke Ver-
ow. If we would just lay them on length of photons passing by and braeck developed a new simulation based
top of each other and interpolate, creates other fascinating effects on the Hollywood film Interstellar. Professor
there is a region in which one (Fig.9). Elmar Eisemann and Professor Alexander
grid does not have any informa- Verbraeck also speak in this episode.
tion to contribute. Without going
18 March 202124.2
Fig 7: Interpolation between grids A and B to find grid I. The grids are warped around the interpolated shadow
to make interpolation possible.
Fig 8: Star effects, from left to right: No effect, Gaussian filter, Trails to make movement go smooth between frames,
diffraction, the image used to make the diffraction.
Fig 9: Effects on brightness and color: From
top to bottom: No effects, Redshift effect, Lensing effect.
March 2021 1924.2 A Physicist who never lost her humanity Aniruddh Kulkarni The epitaph “A physicist who never lost her humanity” is attributed to the respected human being and great physicist Lise Meitner. In the year 1945, Otto Hahn received the Nobel Prize in Chemistry for the discovery of nuclear fission. However, Lise Meitner despite of being associated with the project for the whole of her lifetime, unfortunately did not receive the honor to be the recipient of the Nobel prize. Lise Meitner was born in a Jewish to the field of study still prevailed. collaboration of Lise Meitner and family on November 7th, 1878 in Owing to these restrictions, Lise Otto Hahn started with research Vienna, Austria. The primary ed- could only enter the University on Beta emission. Their collabo- ucation in Austria had been man- of Vienna in the year 1901. With ration prevailed for 30 years, in dated by Empress Maria Theresa Ludwig Boltzmann as her men- which they worked on radioactive in 1775. Till the 1800’s the reason tor, she entered the world of substances and conducted ex- that women are not competent physics. Otto Robert Frisch once periments for obtaining isotopes enough was used as an argument said that Lise Meitner made her of different elements. The results to restrict women from entering vision of physics as battle for the of their work were published in universities. Starting from the ultimate truth, a vision that she three major articles in the year year 1872, women were allowed never lost. Following her doctor- 1908. The World war I started in to graduate yet they remained ate, she moved to Berlin in 1907, 1914 and during this period Hahn excluded from some specific uni- where she attended lectures served in the German gas ser- versities due to the ideology. By by Max Planck. At University of vice headed by Haber, while Lise the year 1901, this ideology was Berlin, Lise Meitner started her served as X-ray nurse for the discarded and women were al- research with radioactive sub- Austrian army. In 1917, they iso- lowed to enter universities how- stances along with Otto Hahn. lated the isotope of the element ever some restrictions pertaining The beginning of the long-lasting protactinium. In 1932, James 20 March 2020
24.2
Chadwick presented a conclu- holm where she took up a post an immense amount of energy is
sive proof on the existence of at Manne Siegbahn’s laboratory. released during the fission re-
neutrons. This proved to be as a This was the place where she action, the idea of using nuclear
stimulus to the research of Lise collaborated with the noted sci- fission reaction as a potential
Meitner and Otto Hahn which in- entist Niels Bohr. However, Lise weapon emerged. The methodol-
vestigation of the effects on neu- and Otto Hahn remained in touch ogy of how an atomic explosion
tron bombardment of nuclei. during that time. In December, would take place, published by
Otto Hahn and Fritz Strassman Frisch-Peierls, became the ba-
In 1934 Ernico Fermi experiment- conducted experiments which sis of the establishment of the
ed with the effects of bombarding was on the process of ‘bursting Manhattan Project. Lise Meitner
neutrons on elements for pro- of uranium’ which later came to however turned down the offer
ducing isotopes. The bombard- be known as the nuclear fission declaring, “I will have nothing to
ment of neutrons of elements reaction. However, the results of do with a bomb!”
yielded isotopes of the element. the experiments turned out to be
However, one anomaly emerged unusual. Lise and Frisch articu- Lise Meitner contributed to the
during this research, which was lated the results using the “liquid foundation of the nuclear fission
the behaviour of element urani- drop model” of the nucleus, that reaction. Between 1924 to 1938,
um. Upon bombarding neutrons the uranium nuclei split into el- Lise Meitner and Otto Hahn were
on uranium there were several ements Barium and Krypton. The nominated for the Nobel Prize
products produced as opposed term nuclear fission was first 19 times, 17 times for the field of
to the definitive results yielded coined by Otto Frisch which ex- chemistry and 2 times for physics.
by other elements. This behav- plained the ‘bursting of uranium’ However, due to the Lise’s emi-
iour could not be explained. This and resulting release of an im- gration and intermittent collabo-
puzzle only grew by the year 1938 mense amount of energy. This re- ration with Otto Hahn, the Nobel
which was a pivotal year for the sult provided a conclusive proof committee did not consider the
scientific community in Austria, that all the elements beyond ura- contribution made by Lise. This
and shortly before World War II. nium in the periodic table could resulted in Otto Hahn winning
not exist in a stable state. The the Nobel prize. This mistake by
In 1938, the annexation of Austria results of the experiment were the Nobel committee was partly
into Germany took place which is presented in Naturwissenschaf- rectified when Lise Meitner, Otto
referred to as the “Anschluss.” ten in 1939. Hahn and Fritz Strassman were
During this political union, many awarded the U.S. Fermi Prize.
scientists were dismissed or The explanation of the empiri- The element “Meitnerium (Atomic
forced to resign. This became a cal results provided by Lise and Number 109)” has been named in
reason for many renowned scien- Frisch had a scintillating impact the honor of Lise Meitner.
tists to emigrate, Lise being one on the scientific community. Once
of them. Lise moved to Stock- there was conclusive proof that
March 2020 2124.2
Master Student Column
Internship at SLAM Ortho
Jan de Jong
Jan is now a second-year master moving to Rotterdam and starting
student in Computer Engineering. my thesis soon, so it’s kind of the
He started his bachelor’s in elec- end of an era.
trical engineering 8 years ago still open, but that’s for another
right here in Delft, during which What do you like to do in your time maybe.
time he was also active in ETV. free time?
I’ve always been interested in What was your motivation for
Tell us something about your- technical stuff since I was a doing an internship?
self! child - being into LEGOs, building During the first year of my mas-
It has been some years since I things with my sisters and stuff. I ter, it was quite hard to find any
have been an active member of really like the technical aspect of interesting courses in the first
this association. So, a brief intro- electrical and mechanical stuff, quarter of the first semester.
duction. I did a lot at the ETV, as but also actually building things Therefore, I decided quite early
well as a board year in my third and working with my hands. I’m on in the year that I wanted to do
year. During my year we also had very happy I could make it come something other than CE cours-
the Lustrum year, so there were together into a study. I also liked es for my free elective space.
a lot of festivities going on. This swimming, and now I’ve picked And, quite thankfully, this op-
year I am still doing the Board of it up again. And of course, be- portunity was presented to me
Studies in Computer Engineering, ing with friends. It was always a during spring/summer: I was ap-
so I’m still a bit involved, but I’m good time at the pub when it was proached by a friend who works
22 March 202124.2
at the company SLAM Orthope- domains of electrical, mechan-
dics, to see if I was interested in ical and computer engineering.
doing an internship with them. But I didn’t really have experi-
ence in building a PCB etc, so
SLAM Ortho is a startup with its that was all new for me. I had to
offices at the YES!Delft incubator learn Fusion360, which is a pro-
on the TU Delft campus. They fo- gram by Autodesk that you can
cus on surgical orthopedic drill- use for fast prototyping for 3D .
ing, and are developing an adapt- I learnt to actually make a board,
er add-on for a surgical drill that although it wasn’t a very hard
can automatically measure the board to make, but it had a good
drill depth in bones. This is used number of electrical components
in plate osteosynthesis wherein a for a first board. There were also
metal plate is fixated to the bones some challenges, and debugging
to stabilize a bone fracture. Their is also a really important skill -
product drastically shortens the very frustrating, but it can also
operation procedure and will be rewarding. There was a lot of
also cause the procedure to be software writing, and although I ciety can be made in the case of
less error prone since the meas- am a computer engineering stu- SLAM Ortho.
urement step is automated. dent, this is one of my least fa-
vourite things. Do you have any advice for cur-
Tell us a bit about your intern- rent students?
ship assignment. It was a really friendly environ- I would recommend everyone to
My internship assignment was to ment. I would like to see that in a do an internship in their master
design and prototype a portable company when I apply for a real programme. It is very interesting
receiver for the sensor adapter job. If you want to really work in a how you can apply the knowledge
that they are currently develop- team and get the best out of each from courses in such an assign-
ing. They had to show it to their other, you need to know each ment. You learn the translational
clients and convince them to buy other - but that’s also one of the step between all the theory, and
it, so it had to be awesome. The things I’ve learned from my board its practical implementation and
whole design loop was walked year. prototyping. Besides this, it also
through and everything, from re- gives a good perspective on the
search into requirements to the What was your favourite part dynamics in a company and how
final assembly, was touched. One about the experience? they differ from academia, so you
of the most exciting things in my It was an amazing experience, don’t end up going to work and
experience was designing and and one of the things I would nev- not know what ‘working’ is, be-
producing a custom SMD PCB. er forget would be the ambience cause you’re in for a treat then.
There is no such opportunity within a startup. The amount of
elsewhere in the CE study pro- freedom that is available to you With all of this being said, I’m
gram. is also quite a large responsibil- very thankful for the opportuni-
ity. I learnt a lot about the design ty I’ve gotten from SLAM Ortho
Within a quarter it was possible cycle, without everything being and the TU Delft. I will soon begin
to do all of these things, and in defined for you in an assignment on my master thesis project and
the end, a proof of concept for by the TU. But because of that will proceed to conclude my time
the portable receiver was deliv- you can let your creativity thrive. here in Delft. And I hope you will
ered. For SLAM Ortho this meant And thus, you can really stay on all consider an internship as a
that a lot of new information was top of next-generation ideas. I part of your master programme.
gained, and further strategies now really know what it is like to
could be researched. actually design electronics for a
company.
Did you feel prepared for it?
I already enjoy building stuff, Besides this awesome ambience,
and also connecting the three a very useful contribution to so-
March 2021 2324.2
Bachelor Student Column
Gijs Lagerweij, Third Year Electrical Engineering
Gijs is a third-year Bachelor stu- it was close to where I live. ly involves theoretical elements
dent electrical engineering at the such as a literature study and
TU Delft. He has done an intern- Why did you choose to do an modelling, as well as practical
ship at Prodrive Technologies for internship for your minor? elements such as design, test-
his minor. There are two reasons for that. ing, performing measurements,
Firstly, the TU Delft did not of- and analyzing the obtained data.
Why did you choose Electrical fer many minors in areas that Another very important aspect
Engineering at the TU Delft? interested me. Secondly, I want- is documentation, which should
To me Electrical Engineering ed to apply the knowledge and help your colleagues understand
was, and still is, a very interest- skills that I have acquired over the outcomes of your project af-
ing and broad discipline with ap- the years. (Work) experience is ter it is finished.
plications in almost every aspect something that goes beyond what
of our lives. I knew pretty early you learn at university, which is Where did you do your intern-
on that I wanted to do Electrical mostly theoretical. An internship ship?
Engineering. It was a combina- helps you develop practical abil- I did my internship at Prodrive
tion of the courses I liked during ities, which is very valuable for Technologies located in Eind-
high school, as well as my hob- your development as an engineer. hoven. Prodrive is a technology
bies. My father, who is also an company comprising over 1300
electrical engineer, was always Do you think there is a large employees and is one of the fast-
happy to help or explain things. difference between studying est-growing companies in Eu-
His support certainly gave me a and doing an internship? rope. Prodrive designs and man-
nudge in the right direction. Definitely. The goal of most cours- ufactures electronics, software,
I chose to pursue this course of es at university is to give you a and mechanics, power electron-
studies at the TU Delft because I broad theoretical understanding ics being the core competence,
liked this university best. I also of various concepts that belong and is active in many different
visited several other universities, to that subject, this is tested at an industries, such as the medical,
but Delft felt like the right choice exam. During an internship, there industrial, automotive and sem-
to me, both in terms of atmos- is no exam, instead, your goal iconductor industries. Their ca-
phere and in terms of the aca- is to complete the project that pabilities include in-house pro-
demic environment. Additionally, is assigned to you. This typical- duction, assembly, testing and
24 March 202124.2
life-cycle-management of elec- had to be analyzed, after which Did the coronavirus have an
tronic products and systems. I had to make a comparison be- impact on your work?
tween the variants. A number Of course, but it was not as bad
Can you tell us more about of custom power modules with as it could have been. Working
your project at Prodrive? integrated passives were also from home is unavoidable these
Within Prodrive, numerous prod- compared against discrete SiC days, but thankfully Prodrive re-
ucts incorporate power modules MOSFETs in identical configura- sponded very well to the virus.
and discrete semiconductor de- tions. They implemented strict rules
vices. These parts are off-the- The results of this investiga- which allowed location critical
shelf products which are typi- tion were quite promising. We work (e.g., lab and production
cally not optimized for electrical observed an improvement of work) to continue in a certain ca-
and thermal performance in the roughly 25% to 50% in steady- pacity. These rules allowed me
specific application. This is es- state thermal performance over to spend a significant amount of
pecially the case for higher-per- the operating range when com- time on-site. Aside from some
formance WBG devices. Prodrive paring the custom power mod- minor delays for ordering PCBs,
is now developing custom pow- ules to the discrete SiC MOSFET components, etc., there was no
er modules using SiC MOSFETs solution. By integrating decou- significant impact on my project.
in an effort to drive down con- pling and snubber into the mod-
verter costs and improve per- ule, we were able to achieve 75% What did you like the most about
formance. These power modules lower undershoot and oscillation doing an internship?
provide a larger degree of flex- than with the decoupling outside I really liked that I was able to
ibility in high-efficiency power the module. completely focus on a subject
conversion products. One way of We have also identified some that interests me. The internship
optimizing the electrical perfor- limitations of the modules and gave me the time required to dive
mance is through the integration potential paths for improvement. into the subject and create some
of various components (snubber, The general expectation is that, good, useful results. Usually, you
decoupling, gate driver, etc.) into with the power semiconductor do not get this amount of time at
the module. technology shifting from Si to SiC university, where you touch upon
The goal of my project was to and GaN, the ‘vertical integration’ many parts of a subject, but never
completely characterize and ver- in terms of packaging will be- really dive deep into the details,
ify the electrical and thermal be- come more and more important at least during the Bachelor.
haviour of various custom power for the company to be able to de- At the same time, I also enjoyed
module configurations incorpo- sign power conversion products the cooperation between disci-
rating passive components. The with high reliability, long lifetime plines. It seems every discipline
assignment involved qualification and low cost/performance ratio. has its own perspective and
of a test setup specifically devel- working together can give rise to
oped to be able to characterize How did you experience the at- interesting discussions and new
both the hard and soft-switching mosphere at Prodrive? insight.
performance of any half-bridge, The atmosphere at Prodrive was
both discrete MOSFETs as well very pleasant. It is quite informal,
as (custom) power modules, de- and you can talk to anyone at any
velopment of a measurement time. My colleagues were happy
procedure, as well as conducting to help and explain anything dur-
the measurements. The electrical ing my internship, either in the
and thermal performance then office or through a Teams call.
March 2021 2524.2
What is Quantum Computing?
Prof.dr.ir. K.L.M. Bertels
Prof. Bertels moved to Portugal
to continue working on the quan-
tum computing challenges. As a
full professor in Porto, he is in
the process of setting up a Eu-
ropean quantum computing re-
search centre to which already
multiple universities or research
groups are participating. The
current members of the initia-
tive are the University of Porto
(P), ENSTA, in Brest (F), CEA in
Grenoble and Paris (F), the Po-
litechnical University in Valencia
(E) and the University of Leuven
(B). The goal is to work on var-
ious accelerator topics and to
expand the tools for the different
layers. QBee is a company that
was created to develop several
quantum accelerators and will be
active at the international level.
To this purpose, QBee will con-
tinue working on the tools and
algorithms that will be needed
once a quantum computational
accelerator can be made.
computer. So, every word in this units (NPU) like Google’s tensor
The content of this article needs article refers to a quantum ac- processing unit (TPU), etc. The
further research, but is based on celerator, which in principle will formal denition of an accelerator
prof. Bertel’s latest research and operate on the quantum mechan- is indeed a co-processor linked
recently published article, and ical principles of superposition to the central processor that is
remains his intellectual property. and entanglement. capable of accelerating the exe-
cution of specific computational
Introduction: Given the recent insights leading intensive kernels, as to speed up
There is growing interest in quan- to e.g. Noisy Intermediate-Scale the overall execution according
tum computing as the new com- Quantum (NISQ) technology as to Amdahl’s law. We now add two
putational direction in which we expressed in [2], we are much classes of quantum accelera-
can keep on building ever strong- more inclined to believe that the tor as additional co-processors.
er computers. In the remainder first industry-based and soci- The rst one is based on quantum
of this article, I will not talk about etally relevant application will be gates and the second is based on
a quantum computer but rather a a hybrid combination of a classi- quantum annealing. The classical
quantum accelerator [1] as up to cal computer and a quantum ac- host processor keeps the control
this point, I have not read a good celerator. It is based on the idea over the total system and dele-
article about what that any end-application contains gates the execution of certain
a quantum computer is. How- multiple computational kernels parts to the available accelera-
ever, what an accelerator is, is and the properties of these parts tors.
quite well understood by the are better executed by a par- Computer architectures have
semiconductor world as almost ticular accelerator which can evolved quite dramatically over
every computational device is be, as shown in Figure 1, either the last couple of decades. The
built on transistors used to con- field-programmable gate arrays first computers that were built
struct a co-processor that will (FPGA), graphics-processing did not have a clear separa-
substantially speedup the overall units (GPU), neural processing tion between compute logic and
26 March 2021You can also read
