We are pleased to introduce Prof. Gerhard Klimeck, one of the speakers at the upcoming IEEE-Nano 2024. He is a professor at Purdue University and vice president for Academic Information Technology and Deputy CIO. His research focuses on computational nanoelectronics, high performance computing, and data analytics. Since 2002, he helped to expand nanoHUB.org to serve over 2 million visitors globally and his team released pioneering scientific computing cloud apps in 2005, which have been widely adopted in academia. His nanoelectronic modeling software (NEMO) established cutting-edge quantum transport modeling, with NEMO5 now used by Intel for advanced transistor designs, leading to numerous publications, high citation impact, and prestigious fellowships and awards.

In this interview, conducted by a committee member of the Publicity team, Minkeun Choi from Pohang University of Science and Technology (POSTECH), Prof. Gerhard Klimeck shares invaluable insights into his career journey, the spirit of overcoming difficult challenges, and the attitude required when collaborating with people.

• Welcome Prof. Klimeck to IEEE Nano 2024. Could you start by sharing with us about your career path and what inspired you to focus on modeling quantum mechanisms?

I left Bochum University in Germany for a 1-year adventure in 1988 to take Electrical Engineering (EE)classes at Purdue and to return with the credits towards my German Engineering Diploma (BS & MSEE combined).   In the spring of 1989, I took a class in Electrical Engineering from Supriyo Datta on Quantum Transport.  I found the topic fascinating, challenging, and by looking at Moore’s law (outside of the class room) future guiding. It was clear 35 years ago, where devices would be heading (at least in my naïve young mind).  Other areas of EE were also mathematical, but seemed too straightforward.  In Semiconductors, I could do the math but I struggled conceptually, understood little, and was challenged.   So, I decided to pursue something hard that might have impact.  

I graduated with my German degree in August 1990 and with my PhD in January 1994.  My advisor Supriyo Datta introduced Non-Equilibrium Green Functions (NEGF) into the field of EE and I worked on resonant tunneling diodes (RTDs) which have nano-scaled heterostructure features in the growth direction and are “infinitely wide” in the plane.   NEGF is the “perfect” method for such devices and I modeled phonon scattering in the device.   However, when you make the lateral dimensions small, then you get Quantum Dots and observe Coulomb Blockade.   I ended up showing Datta that Coulomb Blockade cannot be “simply” modeled in NEGF and developed a rate equation approach that can account for the electron-electron interactions within a dot via many body (Slater Determinant) interactions, tunneling rates due to geometry, and effects of the modes in the leads that connect the dot to the environment.   We modeled individual dots in high bias and coupled dots in the linear regime.  We were really limited by the available computing power.  

Once graduated I started a post-doc with Bill Frensley at UT Dallas in a project at Texas Instruments (TI):  The topic was predictive modeling of resonant tunning diodes (RTDs) where we examined multiple modeling methods.   By 1997 we were able to quantitatively predict RTD device performance, by comparing our NEGF model to literally hundreds of experimental devices.   We even corrected the material grower’s recipes as they were consistently off by 2-atomic monolayers.   NEGF enabled us to include local atomistic orbitals, extended contacts with strong scattering, and realistically extended devices.   Our NEMO (Nanoelectronic Modeling tool) set the standard for 1D quantum transport simulations.

In 1997, I gave a NEMO1D demo with its very powerful graphical user interface to the CTO of NASA/JPL ad a NASA conference.  In 1998, I was hired into the NASA/JPL High Performance Computing (HPC) group to build up a nano modeling team in support of the lab’s microelectronics fab.   Thomas Sterling had just built Beowulf #2 in the HPC group – Hyglac – 15 Pentium Pro 200MHz, 128MB RAM each, 5GB disk each, 100Mbs crossbar, 3.2GFLOPS theoretical total with 80GB disk, and 2GB total RAM. It was a radically different architecture compared to the standard specialized supercomputers such as HP, SGI, or IBM.   It was also about a factor 10-50x less expensive, and one could run for days on it. So, we built NEMO3D for 1 million atom electronic structure calculations in quantum dots – no transport due to the lack of compute power.  The code was highly optimized for low memory use and efficient numerical performance.  Fast forward 6-8 years and Beowulf clusters basically replaced custom supercomputers due to their low cost per floating operations ($/FLOPS). By 2002, effective mass (non atomistic) quantum transport in nanowires/tubes became possible and by 2008, OMEN was developed by Mathieu Luisier at ETH & Purdue for atomistic quantum transport.   We began NEMO5 development subsume all prior NEMO versions in 2008.   In 2015. Intel bought an internal supercomputer that ranked in the top 100 to run NEMO5 for transistor explorations and designs.  In 2018, SILVACO licensed NEMO5 for commercialization.  So it took about 25 years to bring fundamental research in NEGF to industry-ready software.

• It’s amazing that you’ve led the development of various atomistic simulation tools, starting with building the foundations of NEGF modelling and moving on to NEMO1D, NEMO3D, and finally commercialization of NEMO5, which brings it all together. In addition to developing simulation tools, you’ve also been involved in establishing the nanoHub, which has had a huge impact on research, industry, and education around the world, and I’m sure there were many challenges along the way. How have you overcome challenges, and how do you stay motivated and inspired by your work?

At TI, I learned that a powerful GUI and fast simulation delivery is critically important for real tool adoption by experimentalists.  Real use breaks the tool and makes it better.  At JPL, I learned about HPC and tool deployment over the web where we could either reduce compute time or increase the problem size and find ways to share tools.   In 2002, the opportunity emerged to grow nanoHUB from a concept to a national and global cloud platform.   In 2005, we created a cloud before “the cloud” was a thing and we created Apps wrapped around powerful simulation tools, 2 years before the iPhone came out with apps.   We had to build up a completely new software team at Purdue and in 2007, spun HUBzero out of nanoHUB such that other fields of science can benefit from the HUB concept. Michael McLennan was leading the new HUBzero business unit at Purdue with some 20+ FTE.   That made HUB technology development scalable.

For the various NEMO developments, I started out as a lowly post-doc in a group of 5+ full staff at TI, at JPL I built up a group of 4 staff, and at Purdue I lead at some point a group of 35 researchers with 3 research faculty, 1-2 postdocs, and lots of students.   We wrote papers ranging over fundamental physics, quantum computing devices, and applied device designs.  The fundamental physics, the atomistic representations really calibrated the industrial strength tool for modern transistor design.

nanoHUB represented a completely different challenge beyond quantum transport modeling.  Remember in 2002 there was no “cloud” there were no apps.   But how can you enable much larger number of users to use advanced computing on advanced models.   By 2005, we created our own cloud before the “cloud” was a thing and 2 years before we created apps wrapped around sophisticated engines to enable anyone to do quantum dot or MOSFET or other simulations.

• To complete modeling, it is expected that a amount of experimental results will be needed. This work would have to be done in collaboration with many people, so how did you overcome communication barriers to maintain a cohesive team dynamic?

Yes, indeed it is very critical not to do modeling for the sake of model building.  Close connection to experiments has always driven the NEMO tool development.   Often, we were standing second in line to the experimentalists, often to the consternation of the theory students.  But at the end the quality product by a team is important.  Not the individual; just as in soccer one cannot win a game with 11 strikers or 11 goalkeepers…

• If you could give one piece of advice to young scientists starting their careers, what would it be?

Work on the hardest problems that people care about or should care about.

• Finally, what are your expectations for IEEE-NANO 2024? Please give us a brief overview of the tutorial session you will be presenting.

IEEE nano is usually a rather diverse set of presentations, so it should be exciting.   I will give 2 talks on quantum transport and on nanoHUB.   Both presentations will overview some 20+ years of work for the field.

Thank you very much for your valuable time, Professor Klimeck.  

See you in Gijón!