Fly High Engineering

Many companies operate legacy equipment — i.e. obsolete systems that are still operational and reliable but lack modern functionalities and aren’t connected to a modern fully-integrated industrial IT system — experiencing problems like error-prone encoding of production settings and reporting of production outputs, lack of information about process variables that lead to accurate estimates of performance, costs, and schedule, and critical dependence on few key people, who are the only ones with enough experience on such systems.  The last aspect is a very significant source of risk when equipment is 10 or 20 years old — several industries still rely on 30 y equipment — because the machine experts may leave the company and it is very difficult to find new personnel able and willing to learn about how to work with “dinosaurs”.  To mover further, they face decisions between replacement and modernization.

Replacement of old equipment brings enormous advantages — smart machines fully integrated into an Industry 4.0 infrastructure, long lifetime, possibly with higher throughput or new functionalities — but also comes with a significant financial and operational risk.  First, the investment is big and the expected ROI (return of investment) comes after several years, if everything goes as planned.  Unexpected changes in the industry — the first quarter of 2026 has been rich of unpleasant surprises, just to make an example — may disrupt the financial plans and delay the break-even or even make it unattainable.  Second, unless there is additional room for the new equipment to be commissioned while production is ongoing, its installation requires a significant downtime for the removal of the legacy systems, followed by the installation of the new ones and by the verification and validation of all requirements.  Last but not least, the new systems will be very different — much easier, hopefully — to operate.  People manifest resistance to change even when the latter is for the good.  Experienced operators will lose their uniqueness (and the resulting influence) and have little motivation (and perhaps also ability) to learn how to use the modern equipment.  They may even decide to quit the company — a serious problem because their experience is very valuable — without training younger operators.  In short, the disadvantages of a full replacement may (and often do) completely offset the advantages.  The result is a modern and fully functional system, which however does not deliver the expected benefits.  Sometimes this is catastrophic for the business.

Often, the best way to avoid or mitigate these problems is a gradual modernization plan.  Among the many advantages, this brings significant extension of equipment lifetime with reduced downtime risk, improved maintainability, and enablement of ERP/MES integration.  Compared to replacement, modernization interventions have minor costs (often 10-20 times smaller), are distributed in time (as well as the corresponding investments), and can be associated to specific benefits with minor or no disruption on the internal processes.  In particular, the possibility of phased interventions transforms a big disruption into a sequence of minor changes for continuous improvement.

While the advantages of modernization are apparent, one must also say that they don’t come for free.  Modernization is a strategic choice that must be supported by a sound plan for phased improvement.  The most common mistake is to proceed bottom-up, starting from the list of equipment and replacing the oldest units.  Old does not necessarily mean bad.  Replacing things in chronological order is a simple algorithm, and as such seldom is the best option.  Another common mistake is to base decisions on an internally performed OEE (Overall Equipment Effectiveness) analysis.  Unavoidably, the organization experts have biases that don’t allow them to see the picture from a neutral point of view.  If runtime logic and configuration logic are not clearly separated, they won’t notice it because they are used to operate that way.  If downtime causes are not structurally encoded but depend on the experience of the single, which is hard to transfer and document, the root cause will be missed.  If data capture depends on manual entry and human interpretation, then the OEE analysis will simply reproduce existing ambiguity without clearly pointing out what needs to be changed and why.

What is difficult for a modernization process is to understand how all the pieces of the puzzle contribute to today’s business processes, identify the biggest criticalities, outline an organization-aware improvement plan, prioritize interventions based on their expected benefits, ensure that they are consistent with the financial situation and plans, establish how each change will be assessed, plan the work accounting for the needs and aspirations of the relevant stakeholders, execute the modernization steps in a controlled way, and demonstrate that the benefits are actually achieved.  There is no standard recipe for this.  You can’t “buy” modernization the same way you can buy new equipment to replace the old one.

This is why Fly High Engineering starts from the analysis of the business processes and how the people in the organization contribute to them.  Quite often, relatively small updates to the control systems allow to integrate old but reliable equipment into a modern Industry 4.0 infrastructure.  This enables error-free real-time reporting of production data, which in turn allows for precise costing and scheduling estimates and for predictive maintenance, implemented at the higher level to avoid changing what already works reliably.  In many cases, this is enough for the management to get full visibility on what is happening and more control on the planning, with negligible impact on financials and operation.  Another common way of improving how people interact with the equipment is to move to a modern navigation-based HMI (Human-Machine Interface) design, which makes it more intuitive and less error-prone for the operator to interact with the equipment.  Involving the experienced operators in the design process ensures that their know-how is captured.  At the same time, the new HMI is much easier to learn for newcomers and its documentation contributes to the organization’s knowledge base.  When certain components are not only obsolete, but also hard to replace, the risk of disruption is concrete and it may be safer to replace them with modern units capable of replicating their functionalities.  This needs to be judged case by case and the decision is always to be made at the strategic level.  In any case, modernization must start with business goals and demonstrate that technical interventions do deliver the expected benefits.

Fly High Engineering often works on complex PLC-based control systems. A PLC (Programmable Logic Controller) is a computer used to automate industrial processes and manage different operations such as controlling machines, production lines, transport systems, etc.

In addition to the processor and memory, a PLC is typically connected to various units that manage input and output lines, some of which may be dedicated to safety systems (emergency stops, operator safeguards, device protections). In modern systems, the PLC is also interfaced with other units such as user interface panels (HMI), other PLCs (often in master-slave architecture), distributed control systems (SCADA), Manufacturing Execution Systems (MES), and data collection systems (ERP) that provide management with the most important summary information.

The PLC controls devices and manage their input/output

Having to control physical devices, the fundamental task of a PLC system is to manage the signals coming from various sensors, which can represent the state of a switch, the temperature, the pressure, the level of a liquid, etc. When the incoming information is encoded as an analog signal, it is digitized by an ADC (Analog-to-Digital Converter) module. Similarly, a DAC (Digital-to-Analog Converter) is used to convert the digital result computed by the control software into the analog value to be set on the output line, when the connected device needs analog control. To program a PLC, the first step is therefore the hardware configuration, which must correspond to the actual installed devices and their wiring.

The main control cycle includes scanning all inputs and using the values of the various sensors to update the outputs, which drive the connected devices. Depending on the manufacturer, the control software is written in various ways, such as Ladder Logic (a graphical language that represents the various connections in the logic circuit), Structured Text (which is similar to Pascal), Function Block Diagram (a graphical language that represents functions as boxes with inputs and outputs, and the connections between the various boxes), and the more modern TwinCAT C++ from Beckhoff.

Industrial systems require real-time control. This means that the latency, i.e. the time it takes to read all inputs, calculate, and set all outputs, must be predictable and controllable. The latency is usually measured in milliseconds, so we are always talking about very fast processes.

Challenges with PLC systems

The most challenging aspects are often related to the diagnosis of problems in input/output devices, and to handling communication errors and other faults. In particular, the management of the safety system (for example, automatic blocking or switching to a safe action mode if a protection is triggered) requires extreme attention, because problems in industrial systems can not only have disastrous economic consequences but can also cause accidents with deaths and injuries.

One must also keep in mind that there are HMI interfaces for the operator, who must be able to monitor the process and be able to act promptly in case of problems. Often the system defines roles with different privileges, for those who need to set up the production line, for those who just need to keep it running (for example, ensuring that incoming material is always supplied), and for those who need to have access to everything to carry out repairs and maintenance.

Industrial systems exist with various levels of complexity

Today, almost all industrial systems are connected to other systems, using communication protocols such as Ethernet/IP, Modbus, Profibus, etc. PLCs are often integrated with Supervisory Control and Data Acquisition (SCADA) systems and Manufacturing Execution Systems (MES) to provide more advanced control and monitoring of industrial processes.

The advantages of PLC-based systems are their robustness and reliability, being designed to operate in harsh industrial environments, the flexibility they allow, having to be used to program a very wide variety of applications and processes, and their expandability, thanks to the possibility of integrating additional modules to expand the system’s capabilities. However, you also need to consider aspects such as cost, the complexity of their programming approach (they require experts with specialized skills), long-term support from manufacturers, and other strategic aspects. Contact Fly High Engineering for an opinion based on both business and technology aspects.

Fly High Engineering has deep knowledge about Data Acquisition (DAQ) systems, including the trigger systems of very complex and highly reliable instruments, such as the huge underground particle detector ATLAS of the Large Hadron Collider (LHC) at CERN, the Alpha Magnetic Spectrometer (AMS) installed on the International Space Station, and the X-ray imaging spectrometer STIX of the ESA-NASA Solar Orbiter mission.

What is a trigger system?

The goal of the trigger system is to select the “events” that the DAQ system should record, while ignoring uninteresting events, which are usually much more numerous than the interesting ones.  Interesting events are defined by listing several criteria they must meet.  For example, the traffic police use speed cameras to take pictures of cars exceeding the speed limit, which are a minority of all vehicles traveling along that stretch of road.  This is quite a trivial example of trigger system, however.  Most scientific, medical, and industrial DAQ systems must cope with much bigger complexity.  Therefore, it is usually better to think in a sort of reverse logic and focus on the reasons why a certain event should not be recorded.  In the simple example above, the speed camera is sensitive to all vehicles, but it spends most of time ignoring those with speed below a certain threshold.  Similarly, ATLAS must ignore the vastly bigger fraction of uninteresting events caused by multiple proton collisions at each bunch crossing, with 40MHz bunch crossing rate, and save only about 600 events per second (about 10 parts per million).  This number is set by the capacity of the storage built at CERN for this experiment.

The recorded events are then skimmed further by performing a number of “selection cuts” during the offline data analysis step.  For example, hundreds of trillions of events have been collected by ATLAS and CMS during LHC Run 1 from 2010 to 2012, before they could collect few hundreds of events (about 1 part per trillion!), necessary to achieve enough statistical evidence to claim the discovery of the Higgs boson.

Trigger conditions exist of different types

Before addressing such complex systems again, let us review the main types of trigger conditions.  The simplest one is the requirement that a certain value exceeds a predefined threshold, like the speed camera of the police.  Voltage protection is also based on this approach, to make another common example.  The assumption here is that, when the quantity exceeds the threshold, this is caused by some physical phenomenon that we want to study further (e.g., identify the license plate of the speeding car) or prevent (e.g., damage of electrical components).  Sometimes, the direction in which the threshold is passed (increasing or decreasing signal) is also important, as implemented in the edge triggering of oscilloscopes.

On the opposite side, we might be interested in an unbiased, random trigger, with which we can accumulate events that are studied to characterize the environment or situations in which interesting events are supposed to occur.  This trigger may be generated with a clock, for example to acquire events falling in predefined time intervals, random generator, or even a counter (in the assumption that interesting events occur randomly).

Interesting events may also be characterized by specific patterns, like the waveforms that led to the discovery of the first black-hole merge by LIGO, or the four-lepton pattern of the Higgs decay into two Z bosons.  The challenge is how to detect such patterns real-time and with as little contamination as possible from uninteresting events.

Multi-level triggers are necessary to meet very demanding requirements

Complex instruments like particle detectors adopt a multi-level approach to select interesting events.  Traditionally, three trigger levels are identified.  The first level (L1) needs to cope with the highest event rate; therefore, it is based on hardware logics implemented with ASICs and/or FPGAs.  To keep the pace of input events, L1 typically exploits only local data, meaning the chips use input from a single type of detector to take a decision.  The second trigger level (L2) is then used to reduce the event rate further, by combining information coming from different subdetectors.  The third and last trigger level (L3) is finally capable of combining information from all subdetectors to detect the most complex patterns.  Keep in mind that, while the trigger system is processing the event, the detector is “blind”, i.e. it is not sensitive to further events.  This “dead time” ends when the DAQ system is reset, following rejection by the trigger or recording of the event.

The input rate to the ATLAS L1 trigger is the LHC bunch crossing rate (40 MHz) and its output is 100 kHz.  This is further reduced by L2 down to about 1 kHz, which is then halved by the L3 (called “Event Filter” in ATLAS, which groups L2 and L3 into the software-based “High-Level Trigger”).  Random triggers are also employed, to get unbiased estimates of the probability of different classes of events.  ATLAS has a “trigger menu” consisting of several entries, because one may want to record events featuring one or more high-energy electrons, photons, or muons.  Alternatively, a number of “jets” and/or a significant amount of missing transverse-momentum may be required by various physics analyses.  Therefore, a further challenge is how to balance the relative ratios of different “menu entries” with the rigid constraint given by the final event rate on storage.

AMS-02 has only two trigger levels, which would correspond to L1 and L3 of the traditional nomenclature.  The first decision is taken by the Time-Of-Flight (TOF) system, selecting the charged particles crossing the inner bore of the magnet.  A quick veto may be set by the ACC (Anti-Coincidence Counters) and by a lookup table implemented in a FPGA.  At the next level, information from the tracker, calorimeter, and Transition Radiation Detector, is used to refine the decision taken with the TOF and ACC.  The orbit inclination of the International Space Station is 51.6°, which means that the AMS-02 latitude oscillates between 51.6°N and 51.6°S.  Consequently, the flow of particles through it varies, with maximum of 2-3 kHz input rate reached closer to the poles and minimum below 500 Hz (depending on the solar activity) at the equator.  However, the energy spectrum also changes with latitude, with less energetic particles dominating in polar regions, where the Earth magnetic field does not reflect them back to space.   The bandwidth allocated to AMS-02 fixes the maximum rate of recorded events to a level below 100 Hz.  Therefore, the challenge is to make smart use of such bandwidth by recording the most interesting events, like heavy ions (a small minority of all particle flux).  As most cosmic-ray events are protons with relatively low energy, whenever the trigger recognizes this class, a pre-defined suppression factor may be applied by recording only one out of N events with a random choice algorithm.  On the other hand, when the TOF detects a particle with charge 2 (a helium nucleus) or higher, there is no further suppression by the other trigger level.

Summary

In summary, a trigger system is always necessary to start the data acquisition process.  The design of the trigger is the result of a trade-off between different requirements and constraints and is one of the most delicate aspects of the experiment design.  The goal is to collect as many interesting as possible with the little fraction of “background” events compatible with the data acquisition capacity.  Get in touch with Fly High Engineering to know more.

Fly High Engineering typically gets involved into problems where the technical challenges are important. However, the biggest issues often arise from the people, rather than from the technology. In particular, they may come from bad communication, leadership with a hidden agenda, and solutions to false problems. Therefore, these aspects play a fundamental role in all projects, whether technical or not.

Trust comes from communication and leadership

Modern history offers plenty of examples of strong leaders, who could move the masses in some direction, either “left” or “right”. Their followers completely trusted them and, upon their guide, acted in decisive ways, sometimes for the good of everybody, sometimes not. What is the “threshold” for a leader to overcome, to gain such a trust? Which methods allow leaders to gather engaged followers? While it’s difficult to provide general answers to these questions, something can be learned by observing how follower react to leaders adopting different communication styles.

Leaders and followers share communication styles

A given communication style “pairs” leaders and followers, in the sense that a leader adopting that style naturally gathers around people who want to communicate that way. For example, people who want to debate on every detail won’t easily digest commander-like leaders, who just tell them what to do. On the other side people, who just want trusted leaders to tell them what to do, don’t react warmly to someone who wants to engage them into detailed discussions.

Prompt and delayed actions are generated by simple and detailed communication, respectively

In recent times we had examples of very “vocal” political leaders, very effective in moving masses. They go straight on every problem, imposing their view and telling people what to do. Apparently, their followers are keen to act promptly upon these calls. On the contrary, their opponents seem to be less effective because they “lose time” in articulating more detailed positions, because this is what their followers expect them to do. Consequently, when the first have executive power they produce several prompt actions, while the others proceed more cautiously.

Prompt and delayed actions may eventually lead to the same future

In practice, one leadership style leads to prompt actions that have a chance to solve urgent problems, while the other naturally produces a lower throughput of actions, almost preventing prompt reactions to urgent issues. Actually, as the human capacity to predict the is quite little future (if any), neither of the two processes ensures that the taken decisions will bring any improvement. Very often, in practice the first approach produces a series of sharp changes that, when considered in perspective, may look quite noisy; the other approach produces less apparent variability, but this is because it makes less actions per unit time. External and internal constraints often induce a long-term “drift” that is essentially the same, even when left and right governments alternate. This is simply because some changes are unavoidable for a society to survive in the changing environment. Some changes are however so rapid to make a natural selection on the communication style.

Acting carefully won’t bring you out of a fire

If the majority of people perceive a problem as particularly critical and urgent, only the direct and perhaps simplistic communication style of commander-style leaders has a chance to find a suitable solution. Perhaps we are in times where this is the case, as it is suggested by the increasing polarisation of the public discussions. “Polarisation” simply means that people want to be told whether they should choose white or black and expect prompt actions, rather than detailed discussions with which they can certainly understand better the situation, unfortunately too late for the solution to be perceived as effective. Therefore, people prefer betting on head or tails, hoping to get at least 50% probability of success. Sadly, often this means only one thing: 100% probability of failure. We all know about this. Simply, we don’t like being told about it.

The leader’s agenda

If we cannot predict the future and the long-term trend depends only in a minor way on the political side of the leaders, why do we need leaders at all? And how can such people gain so much success and followers?

Leaders are necessary

Leaders are always welcome because, once accepted as guidance, they let us save mental energy by telling us what to do. This is an instinctive attitude that we all have, and many other living beings share with us too. I can’t see a world without leaders, unless something else — an all-mighty AI? — tells us what is best for ourselves. Therefore, let’s address the other question. The answer is that leaders become effective when their agenda is promoted in such a way that their followers see some advantage for themselves.

Communication is driven by the leader’s agenda and by the followers’ interests

Thanks to crafted communication and by leveraging on the needs of the followers, leaders get more and more influential. In practice, it may (and it does) happen that leaders don’t solve the followers’ problems, but nevertheless advance in their career and bring forward their own agenda. The latter is very often hidden to the followers, who only care about their own issues. They are happy about the promises they hear from the leader and are ready to excuse her/him for failing to address their problems, because (1) their leader shows — but perhaps does not sincerely possess — sincere interest and attention to them, and because (2) their leader admits that she/he, as everybody else, could not be able to predict what actually happened and still promises to continue striving to provide a good future for the followers.

Influence and power may go along or may not

Note that the influence of leaders may be paralleled by their effective power on the followers, as it is the case for the management in an organisation, or may not, as it is the case for the opinion leaders in the same organisation, who can influence the decisions of their managers. Therefore, real-life situations may be quite tricky to understand, also because there are several leaders acting at the same time, whose agendas partially overlap, despite their attempts to look different. The main difference is that influencers who don’t have a powerful role in the organisation need to gain a lot more followers’ support to become effective, compared to those sitting already in a powerful position.

The solution must be compatible both with the leader’s agenda and with the followers’ needs

Many projects failed because different stakeholders were pursuing incompatible objectives. The only way to prevent failure is to realize what is the root cause, often the most difficult step, and then act to solve or mitigate the backstage issues that have been uncovered. This is by far not an easy task, as difficult decisions have to be taken.

Three steps to rescue the project

1. Understand the motivations of all stakeholders. They won't tell you about their hidden agenda! Therefore, this takes quite some amount of inference and emotional intelligence, supported by the capability of finding information and hints that suggest what they don't want to tell you.
2. Decide which motivations are compatible with the sought solution. Assess if the overall influence of the stakeholders pushed by such motivations is big enough for the project to have a chance of success. You can't make them all happy! It's impossible, it's not your fault.
3. Decide whether you'd better to abort the project or to push some of the stakeholders out of the game. Both options are painful, and both are fair. Consider strategic aspects and their impact on your own agenda before taking a decision. What pays more on the long term? Can you survive long enough to reap such benefits, or shouldn't you instead go for the short-term advantage and hope that you'll find some way of avoiding the long-term loss?

People don’t like change, not even when it’s for the better

Don’t forget this simple but very important thing! We don’t like changing our habits. Therefore, we oppose resistance even to changes that aim at simplifying our life. Moreover, if the changes are imposed by others, we oppose by default even before thinking about pros and cons. Overcoming the resistance to change is often difficult and may mine the success of what we are doing. The only possibility to mitigate the problem is to (1) analyse all the relevant aspects since the beginning, and (2) involve as many stakeholders as possible in the formulation of the plan. Although change is always difficult, we have more chance to make it if it comes from our own decision.  :-) 

Summary

Often the biggest difficulties in any project arise when the identified solution is not compatible with the interests and needs of all stakeholders. Moreover, the resistance to change adds further obstacles for the success of the project. The project success is much more probable if we can perform good identification of influencers, power-gamers, followers, and understand their needs and, if possible, goals. No matter if they tell you that the problem is technical, there is always a human side that deserves attention and due consideration. Solving the technical problem and leaving people unhappy is a failure, not a success. Therefore, you’ll have to become one of the leaders, pay attention to your communication style and adapt it to different stakeholders. Find a solution that brings benefits to all (or to most of them) and make sure they understand the expected benefits, even before starting to do any technical work!

Quantum computing is an ongoing revolution and promises breakthroughs in many fields, like the ability to find the geometry of complex molecules, which is expected to boost fields like biology and pharmacy, and the factorization of large primes, which threatens the security of our confidential information.   Such an important field relies on the availability of quantum computers, which I’m going to introduce in the simplest way as possible.   A more complete presentation of the subject is provided here.

First, let’s clarify the name.   A modern computer is in the essence a calculator operating with the binary system and we heavily rely on it because it is much more precise and fast than what we can do ourselves.   “Binary system” means that a computer is built with components capable of recognizing and manipulating symbols in a set containing only two elements, which are called “false” or “zero” (0), and “true” or “one” (1).   The reason for choosing the binary system is that it ensures the maximum reliability when implemented with electronics, in the sense that there is the maximum difference between the voltages representing all possible states, minimizing the probability of misunderstanding each single state.   As a side effect, arithmetic and Boolean logics can be implemented with the same circuits.   With computers, everything is encoded in sequences of zeros and ones, stored as sequences of information units called “bits”, each assuming value 0 or 1 depending on the computation.   Normally we group sets of bits in bytes (1 B = 8 consecutive bits) or some multiple of them, like the Kilobyte (1 KB = 1024 B = 8192 consecutive bits).   A computer is a state machine that, at each cycle, possesses a well-defined set of binary values in each bit used by the algorithm.

On the other hand, quantum computers are based on qubits, which not only can represent the two logical states mentioned above, but can also exist in the so-called quantum superposition of such states.   This is a unique phenomenon, completely absent in our daily life, characterizing quantum systems, i.e. the microscopic entities obeying Quantum Physics.   If we consider two qubits produced to be in the superposition of all four possible combinations of 0 and 1, a measurement of their status will end up with a single configuration, among 00, 01, 10, and 10.   By repeating many times the measurement on identical two-qubit systems, the relative frequency of each possible state will tell us what is its probability.   An important step when designing quantum algorithms is indeed to ensure that there is a clear preference for the “right” or “optimal” result, such that one does not need to perform many repeated measurements to find the desired solution.

Nature makes it all the time.   For example, the protein geometry in three-dimensional (3D) space is formed quickly and spontaneously while the amino acids are sequentially added by   ribosomes using the pre-messenger RNA (mRNA) as template.   In practice, the 3D geometry of a protein minimizes the Gibbs free energy, therefore a classical algorithm to compute protein folding, could be the following.   Place the first amino acid in the origin of the reference system and compute its electrostatic field in the surrounding space.   Next, add the second amino acid and place it such that the chemical bound with the previous one minimizes the Gibbs free energy of the resulting system.   Then compute the electrostatic field of the system and add a third amono acid.   At this stage, you have to consider all possible chemical bounds of the third unit with each of the previous ones, allowing also for a reconfiguration of the geometry with the goal of finding the configuration with minimal Gibbs free energy.   This problem is obviously more difficult to solve than the previous step, because there are more possible choices for the overall configuration.   Next, you add the fourth amino acid and repeat the process, this time with many more possible choices.   You easily understand that the computational complexity of this algorithm makes it unable to address long sequence of amino acids.

If the problem is so complex, how can proteins be synthetized by our cells?   The answer is that Nature does not apply a classical sequential algorithm like the one described above but exploits Quantum Physics to “probe” all possible solutions at the same time.   If we have a quantum computer that can mimic this process with a suitable algorithm, then we can compute the 3D geometry of proteins in minutes, while the most powerful computer would take a time longer than the life of the universe to complete.   Biochemistry, which is the set of chemical reactions related to living beings, heavily depends on the 3D structure of proteins.   Therefore, the latter is a very important field of study, for example in pharmacy, and you can easily imagine the high expectations on quantum computing on this side.

Of course, building quantum computers is not easy.   The reason is not that we don’t understand them, but rather that they have a fundamental weakpoint: Quantum superposition allows a small set of bits to “sample” a huge parameter space, but is also very fragile.   Any interaction with the environment, whether desired (a measurement) or undesired (e.g. thermal fluctuations), will break the superposition and make the qubits system “collapse” onto one of the possible states.   The main issue is that, once we prepare the initial configuration of qubits in a suitable superposition of states, random fluctuations in their environment (which cannot be completely isolated from the outside world) sooner or later will destroy this superposition.   The problem is known as decoherence and every quantum computer is characterized by a typical decoherence time, after which the carefully prepared state is lost.   Hence, a quantum computer must be able to configure the desired initial superposition and then run the quantum algorithm in a time sufficiently shorter than the decoherence time to ensure successful completion of the task.   This is not trivial at all.

The best way of increasing the decoherence time is to improve the decoupling of the qubits from the external world, and the big enemy is heat.   The first commercial quantum computer was the IBM Quantum System One and IBM is engaged into a challenging 15-year quantum roadmap including a full technology stack and offering for free 10 minutes per month of quantum computing time on a 100 qubit system.   IBM is investing a lot on research about cryogenic electronics for quantum computing because of the very low temperature required by its qubits to minimize the effects of thermal fluctuations.

In addition to reducing unwanted interactions with the external world, quantum computers adopt error correction to detect and correct single qubit errors.   This requires entangling multiple qubits to store and protect a single logical qubit’s state.   In practice, for each “useful” qubit one must actually implement a system of multiple entangled qubits, which means that the actual number of qubits is significantly larger than the declared number of usable qubits.   The lower limit for error correction is 5 qubits, but common algorithms employ 7 or 9 qubits to encode a single usable qubit.   A performance comparison of different approaches is provided by a recent Nature article written by Google Quantum AI and collaborators.

Quantum entanglement is a property that sets certain quantum systems far apart from macroscopic systems.   When two quantum states are entangled, a measurement performed on one of them not only “collapses” the previous superposition of states into a specific quantum state, but also immediately tells us what is the state of the other entangled state.   For example, the gamma-rays emitted back-to-back in a positron-electron annihilation process, widely used in Positron Emission Tomography (PET), are very energetic photons whose polarization is strongly correlated, because they are emitted as an entangled state.   This may be used to improve the signal to noise ratio in PET reconstruction, as explained by Watts et al. in their Nature paper.   What matters here is that a measurement of the polarization of one photon tells us also what is the most likely polarization of the other, without the need of a second measurement.

This property of entangled states is widely used in quantum computing.   First, it is used to encode quantum information in a redundant way on a set of entangled qubits.   Second, if one qubits experiences an error like a bit flip, the entanglement ensures that the system as a whole can detect the problem without the need of measuring the qubit directly.   Finally, in such a case the system uses entanglement to correct the error, often using majority voting.   Therefore, entanglement is a critical ingredient of quantum error correction.

A very simple scheme for error detection and correction, which is commonly adopted in critical systems like the space missions on which I worked for 20 years (including AMS-02 and Solar Orbiter), is to add to each byte two more bits, called parity bit and CRC, whose values are known for the true representation of the byte.  This increases space occupancy by 25% (from 8 bits to 10 bits) but it allows the underlying hardware to perform realtime error detection and correction, which is a big advantage in noisy systems.  In addition, mission-critical components add one more layer of redundancy by employing majority voting, which means that three identical copies of the same code and data exist and the decisions are taken based on the values that appear at least twice among the three copies.  In all cases, the main goal in error correction is to reduce the probability of a mistake to negligible levels, with respect to the expected amount of computation.

Assuming that the random bit flips are due to uncorrelated phenomena, the probability of two bit flips is the product of their single bit flip probabilities.   As a probability is always a number between zero (the impossible event) and one (the certain event), the product of two probabilities cannot be larger than the individual probabilities.   If the latter are low, a double failure has a neglible probability and can be neglected.   Think for example about single bit-flip probability to 10^–6 (one part per million), which gives double bit-flip probability of 10^–12 (one part per thousand billion or per trillion).   This reasoning applies both to classical and quantum bits of course.

Quantum error correction also aims at reducing the probability of unrecoverable errors.   Similarly to ordinary error correction, redundancy is employed to “distribute” the information across multiple qubits.   However, quantum error correction must be obtained without performing any measurement, otherwise this would destroy the carefully prepared quantum superposition.   This is done with smart combinations of entangled qubits, such that a spontaneous bit flip in one of them would be “healed” automatically by the entanglement itself, without external intervention on the system.   Of course, this is only possible if the rate of errors is low enough, such that there is at most one “wrong” qubits in the entangled state (this is known as the threshold theorem).   Compared to classical error correction, which requires a relatively small number of extra bits, quantum error correction requires significantly more “space”.   The most famous error correction approach is probably the one proposed in 1995 by Peter Shor, which employs 9 physical qubits to obtain a single logical (i.e. entangled) qubit, capable of automatically correcting any kind of single-qubit error.   The extra space is essentially one order of magnitude greater than the number of usable qubits!   This is one of the reasons why it’s so difficult to obtain quantum computer with many (usable) qubits.

All promising approaches to quantum computing nowadays rely on sets of entangled qubits providing a reliable subset of logical qubits, therefore bringing part of the algorithm into the architecture of the quantum processor itself.   Among such topological structures, Microsoft has recently announced the Majorana 1 processor, which could boost scale-up in the next years and win over the competing approaches.   This chip explits Majorana fermions that can be produced and manipulated on a new class of materials, which Microsoft calls “topoconductors”, which exhibit a new state of matter.   This was actually found when measuring the properties of semiconductor–superconductor device, which provide high-mobility two-dimensional electron gases as documented in this (very technical) 2023 review.

A fermion is a particle that follows Fermi-Dirac statistics, named after Enrico Fermi and Paul Dirac.   Elementary particles like electrons, protons and neutrons are fermions.   Moreover, the fundamental particles that in the Standard Model of particle Physics are the building blocks of all forms of matter are all fermions.   They are all characterized by one important property: Identical fermions cannot occupy the same quantum state.   There is another class of particles, called bosons, which include photons and the other particles mediating fundamental interactions, with a diametrically opposite behaviour: Any number of identical bosons can occupy the same quantum state. They obey the Bose-Einstein statistics, named after Satyendra Nath Bose and Albert Einstein.

Fermions and bosons are characterized by half-integer and integer spin, respectively.   We call spin the intrinsic angular momentum of a particle, distinct from its motion in space (which may be associate with its own angular momentum).   Elementary particles, like those mentioned above, possess a tiny intrinsic angular momentum which, measured in units of the reduced Planck constant, assumes integer values like 0, ±1, ±2, etc. for bosons, and half-integer values like ±1/2, ±3/2, etc. for fermions.   Systems composed by elementary particles may also have spin, which is the (vector) sum of the individual spins.

When we provide enough energy, we can create new matter in the form of particle-antiparticle pairs.   This happens billions of times in particle accelerators and demonstrates with incredible precision the validity of the energy-matter equivalence implied in the most famous equation by Einstein: E = m c².   Here, the constant c is the speed of light in vacuo, and its square is a huge number (just a bit below 10¹⁷ m²/s²).   This means that a small mass is converted into huge energy and vice versa.   Particle and antiparticle (for example, electron and positron) must have opposite quantum numbers, such that they can disappear completely when they meet (this is called annihilation).   Therefore, the must have opposite electrical charge or be neutral, and their spin is opposite too.   Hence, all fundamental fermionic particles differ from their antiparticles.

However, Mathematics tells us that it does not need to be so.   In 1937 Ettore Majorana found that a relativistic wave equation may be written in such a way that only real numbers are needed. Quantum Mechanics heavily relies on complex numbers instead, and particles and antiparticles are described by complex conjugate wave functions.   In contrast, solutions of the Majorana equation are fermions that coincide with their antiparticles, called Majorana fermions.   Although no known elementary particle is like them, there are physical phenomena, called quasiparticles, that behave like particles but are not. They are the collective behaviour of a set of particles or
vibrations. One possible realization of Majorana fermions is exploited by the Majorana 1 processor, to achieve fault-tolerant quantum computation (more details can be found in this article).   This processor could boost scale-up in the next years and win over the competing approaches.

Are we so close to the epoch in which quantum computation becomes able to address biochemistry problems?   There is quite some excitement about it in fields like pharmacy, where big investments are made each year to discover new drugs.   While this is really promising, there is also a different form excitement arising from the threat that quantum computing poses on our current cryptography.  
As most current encryption approaches (like RSA) depends on the difficulty of factoring large numbers, quantum algorithms like Shor’s algorithm can break cybersecurity in minutes.   If you consider the amount of confidential information and communications currently stored worldwide over the Internet will be soon vulnerable to this kind of cyberattack, you’ll certainly understand the worries by governments and other organizations about quantum computing.   Anyway, Shor’s factorization algorithm is brilliant and fascinating and you can get a wonderful explanation of how it works in this video by Veritasium and have fun 🙂 while you learn how to adopt new encryption algorithms that enable quantum-resistant communication.