Battery management systems are growing increasingly smarter with innovations in software and hardware that enable more accurate estimation of battery state of charge and health, along with predictive diagnostics.
Key Takeaways
Batteries are becoming more pervasive and important with the proliferation of electric vehicles, robots, drones, and aircraft. But as consumer concerns for safety and vehicle range increase, manufacturers are considering a switch from a nickel-based lithium-ion (Li-ion) battery to a lithium iron phosphate (LFP) battery, which is also cheaper, longer-lasting, with better mineral accessibility.
As battery chemistry advances, so does the technology behind battery management systems (BMS). Modern BMS electronics are now making use of electrochemical impedance spectroscopy (EIS), digital twins, and embedded artificial intelligence (AI) to analyze impedance more thoroughly. This improved understanding allows for more accurate predictions of a battery’s state of charge (SOC) and state of health (SOH). Additionally, each unique battery chemistry calls for its own balancing methods. Altogether, these innovations have ushered in a new era of the software-defined battery.
In the past, many BMS algorithms were based on lookup tables (LUTs). “These do a good job, but come with certain limitations,” said Puneet Sinha, senior director and global head of battery industry at Siemens EDA. “We are now seeing the possibility of moving away from those LUT-based rigid systems. Customers are asking, ‘Can you do impedance and electrochemical impedance in a more online way? Can you do more advanced diagnostics?’ To do that, you need the right algorithms, and they need to run either on board in the BMS or somewhere in the cloud that the BMS can interact with. It is quite exciting and at an early stage, but there is talk of running AI models on BMS, so that rather than relying on lookup tables, you can run an executable digital twin on-device, giving a better, more accurate estimation of certain things.”
An AI-enabled chip — typically an MCU — can run neural network models on the BMS to do some diagnosis. “These can help to be more accurate in terms of estimation of state, but also become more predictive, especially with EVs,” said Sinha. “‘Something bad is about to happen. Can you pull in for servicing or warranty repair?’ This has always been a desire, but in the past, there were certain limitations, such as not having some of those AI capabilities.”
Finding new ways to accurately estimate the state of charge and state of health is especially critical for LFP battery chemistry because the cell voltages show a relatively flat state of charge curve, making it difficult to calculate usable energy. “One of the issues with LFP cells is that their voltage is very flat, so even very minute errors may build up over time, leading to a 5% to 10% loss of energy,” said Sinha. “Companies are looking to invest in new, advanced diagnostic methodologies to solve this. EIS is one of the ways.”
Others agree that the technology is in flux. “The field is moving from slow, lab‑only impedance measurements toward fast embedded measurements that operate alongside existing passive and active battery balancing, and battery management system technology,” said Masoud Rostami-Angas, lead consulting engineer at Synopsys. “Full‑spectrum electrochemical impedance spectroscopy is still the most precise impedance measurement method for mechanism-level insight, which is essential for mechanism‑level degradation and failure analysis. We’ve recently done extensive lab work on testing on very low (milliohm level) impedance cells and have been able to achieve accurate, reliable, and highly repeatable EIS results. While the test results are very accurate, the testing is not quick. Depending on the frequency range, it may take 15 minutes to 2 hours.”
Practical, high‑fidelity alternatives are evolving. “Other approaches include targeted frequency sweeps, multi-sine excitation, PRBS/broadband perturbations, and high‑resolution current‑interrupt methods,” noted Rostami-Angas. “These could be used to estimate impedance during normal cycling with acceptable accuracy, but much more quickly.”
Accurately measuring battery cell impedance, or resistance, is critical to estimating the pack’s current state of health relative to its end of life (EOL). “It enables accurate quality assessment, supports forecasting of remaining useful life (RUL), and helps reduce expensive warranty‑related costs,” said Bryan Kelly, principal engineer at Synopsys. “It’s worth noting that impedance is not the same as resistance, even though the terms are often used interchangeably.”
Internal resistance testing measures the opposition to direct current (DC) signals, resulting in a simplified version of a battery’s true behavior, according to a blog by Christian Loew, product manager in Keysight’s Automotive and Energy Solutions group. “Impedance testing measures the opposition to alternating current (AC) signals at various frequencies for a more comprehensive battery profile. Resistance is the first approximation of the battery’s behavior, while impedance with frequency dependence can provide a more accurate picture.”
Measuring cell impedance is similar to putting a stethoscope on the heart of the battery and getting an EKG from every one of the cells in every battery, explained Clint O’Conner, co-founder of True Balancing. “Impedance is like the multi-dimensional cousin of resistance,” he said. “It’s a vector in the imaginary plane. It has resistance and reactance of the components to it, and it has a frequency associated with it. EIS gets impedance data at multiple frequencies, typically ranging from 1 kilohertz down to something less than a hertz. The dream of everybody who makes a battery is, ‘Give me impedance data, real time, from the field.’”
A cell management or monitoring unit (CMU) in the BMS keeps track of and manages groups of individual cells, typically between 12 and 20 cells, getting data on the voltage of each cell, capturing data from temperature sensors, and performing balancing, said O’Conner. A firmware change to the CMU also can enable it to capture impedance data. By looking at patterns, trends, and anomalies, and comparing them to a master production record, impedance data can provide information about why and how batteries are failing, and help make better batteries. “It can give you early warning of cell failure, in particular thermal runaway and risk of battery fire.”
When the data indicates that derating thresholds are being approached, the BMS can notify an external ECU to alert the user. “A single impedance measurement of a cell is not a reliable indicator of battery state of health,” Kelly noted. “Rather, it is the change in impedance, which correlates with shifts of cells’ internal bulk resistance, that provides meaningful insight into a pack’s overall SOH and how long it can continue to perform within the user’s operating environment and usage patterns.”
Still, there’s work to do to bring impedance measurement down to a cost where it can be embedded into EVs. “This is really hard to do, primarily because you need to excite the battery,” O’Conner said. “You put in a current and look at how the voltage of the cells changes in response to the current, or excitation current and voltage response. Putting in a sine wave is the standard approach for exciting the battery. You put in a sine wave current, you get a sine wave voltage back, look at the response voltage, and compare. Then you get impedance data. But it turns out that generating an accurate sine wave is expensive.”
BMS modeling, verification, and validation
Battery and energy management are a massive part of the success of a given EV platform because there can be the same amount of battery, but not the same range. “We are seeing a lot of companies looking at model-based system engineering, and digital twins are a big part of how they have to optimize, not just every component, but how all of these components need to work together,” said Siemens’ Sinha.
As advancements in battery management strategies continue, it becomes equally important to consider the practical challenges of testing and evaluating these systems. This shift in focus underscores the need for robust methods that ensure reliability and performance under real-world conditions.
Here, physical test‑bench evaluation of battery systems remains challenging.
“Many critical scenarios — especially those spanning wide voltage, current, and temperature ranges — against wide variation of loads, simply cannot be exercised exhaustively on a testbench, or mule platform,” said Synopsys’ Kelly. “Only through a highly accurate virtual prototype can the full spectrum of BMS attributes be explored. In this context, the accuracy of the underlying models and the robustness of the simulator become essential. They determine whether the virtual prototype environment truly functions as an executable specification — one that engineers can rely on with confidence before moving to physical integration.”
Mathematical models based on cell voltages, pack temperature sensors, and current sensors are used to estimate the state of charge of the whole battery. Digital twins can go further and develop a mathematical model that tries to replicate, as exactly as possible, the real-life function of the battery. “You’re taking data from your battery, pumping it into your digital twin, and trying to model the health of your battery,” True Balancing’s O’Conner said. “If you can provide impedance data on each cell, that will greatly improve the accuracy of all of those models and reduce the processing power required to run them. The math won’t have to be as extensive because you’re going to get a really big parameter input that increases your accuracy quite a bit. That’s one of the reasons why getting impedance data is such a holy grail.”
These modeling and simulation advancements highlight the intricate interplay between battery management and performance, yet their effectiveness ultimately depends on how well the batteries themselves are maintained and balanced. This leads to a critical aspect of battery management systems, battery balancing, which ensures that every cell operates efficiently and contributes to the overall longevity and reliability of electric vehicles.
Battery balancing
The BMS typically performs passive balancing with low-cost MOSFETs, but active balancing is gaining ground. “Design applications and their implementations continually evolve, and different manufacturers may choose to implement BMS architecture in different ways,” said Synopsys’ Kelly. “These decisions typically reflect tradeoffs between functionality, reliability, and cost. For example, while certain approaches — such as passive cell-to-cell balancing — are less efficient, they remain widely used because they are simple, inexpensive, and reliable to implement. Cell balancing during charging is only one element of a much broader BMS architecture, which must coordinate many other monitoring, control, and safety functions across the entire battery pack.”
The BMS is very active during the stage when power is coming in. “It’s monitoring cell voltages and temperatures within the cell pack, and it’s able to do balancing to make sure that the cells are being charged easily,” said Jim Pawloski, director of applications engineering at Infineon Technologies. “That’s a very big part.”
Fig. 1: Automotive battery cell monitoring & balancing. Source: Infineon
Cell-to-cell balancing is typically performed when the vehicle is parked overnight with the on-board charger active. “The BMS can capture a performance snapshot of each cell for long-term tracking,” said Kelly. “By applying a controlled charging sequence, the system can measure voltage, current, temperature, and charge time (delta) data, calculate internal bulk resistance, and store these results in memory. This diagnostic sequence can be executed daily, weekly, or monthly.”
Batteries in an EV are connected in two ways for different purposes — in parallel for added capacity, and in series for higher voltage. When batteries are connected in parallel, all the battery cells function like one big individual cell and don’t need balancing. “They tend to rise and fall together,” O’Conner said. “You have all those individual tanks, but there’s a little pipe connecting all of them at the bottom. If one is draining a little faster than another, they’re going to tend to level out because of all the little pipes across the bottom connecting them.”
Battery balancing is needed for in-series battery cells, which are connected from the positive terminal to the negative terminal, enabling a higher electromotive force. An EV battery can have anywhere between 100 and 400 cells connected in series, and that number is growing. “If you’re not balancing, when the first cell reaches full charge, you have to stop, because if you charge any more, you’d be damaging that cell,” said O’Conner. “That means between 99 and 399 other cells didn’t get fully charged. Then you’re driving the car and discharging the battery, and when the first, weakest cell, hits bottom, you’ve got to stop, but there are between 99 and 399 other cells in the battery that have energy in them that you didn’t get to touch.”
Active balancing is more intelligent, but more expensive than passive balancing. “Instead of draining the battery, bringing everything down to the lowest common denominator, active balancing moves energy from cell to cell,” said O’Conner. “You’re keeping the energy in the battery, except for a little. Anytime you move energy, there’s going to be some waste.”
Since battery use is scaling quickly and shifting toward LFP chemistry, there is a growing need for active balancing technology that is low-cost, efficient, and effective. “There’s almost a panic for that now because of the shift from lithium-ion batteries in big energy storage, such as the giant batteries that store energy from solar farms and wind farms,” said O’Conner. “Almost all of them are LFP now. In commercial and industrial vehicles, delivery vehicles, buses, and construction vehicles that are going electric, almost all of them are LFP, too.”
Reliability, safety, and security
At the same time, innovations in hardware and software will help improve the accurate estimation of the state and prediction for the future state, which improves overall reliability. “There is tremendous value in that, not only for EVs, but increasingly more for battery energy storage systems (ESS),” said Siemens’ Sinha.
Energy storage systems must be designed to withstand being in the field for 20 years, especially at the utility scale or in AI data centers. There are different needs for companies that are providing ESS containers versus the operators of the data centers. “The container is going to operate for X number of years, and they need to minimize your servicing costs, so there is a lot of warranty cost that comes into play,” said Sinha. “For the operator side, if they don’t have an accurate estimation, both in terms of the state, but also what’s about to happen with their battery, it can lead to a lot of revenue loss, especially at utility scale. This is critical, especially for AI data centers. The last thing you want is your energy storage to go down, because it can have severe implications.”
In the data center, batteries, battery management systems, energy storage systems, and energy management systems work together to ensure a robust uninterruptible power supply (UPS) system.
“In the data center, even a brief outage can disrupt large-scale AI training runs that take days or weeks to complete,” said Piero Bianco, senior director of product marketing for chips at Rambus. “The UPS acts as the first line of defense that preserves data integrity and protects the power chain from sudden disturbances. It also gives the broader power infrastructure time to manage controlled shutdowns. Without a dependable UPS layer, the entire stack becomes vulnerable to expensive downtime.”
Fire is another concern for both EVs and data centers. Safety systems can be tied in with security controls to avoid such events.
“As batteries age, they need appropriate security monitoring,” said Dana Neustadter, senior director of product management at Synopsys. “If they had more hardware-based security to cryptographically enforce the life cycle limits checks, then there would be some form of secure counters to check that the lifecycle of that battery is approaching a dangerous state. This can be achieved by putting specific interlocks in the system and sensors to monitor the temperature or the thermal state, along with leveraging signatures for the service tools, to make sure that you know the systems are serviced correctly by the people who are supposed to service them. It’s going to move more toward solutions at the hardware level, not just software control. It must be access control driven by hardware.”
Future outlook
Effective battery management and balancing are growing in importance, with techniques continuing to evolve along with battery use cases and chemistries. “These are becoming massive business drivers, where you’re seeing that BMS is evolving from embedded lookup tables to the software-defined battery,” said Siemens’ Sinha. “Companies are coming up with innovations and new technologies, both on the hardware side, at the board level or chip level, and also on the software side.”
Considering all these new, advanced features, will EV batteries get to a point where they can last for a week?
“The first answer is yes, the second answer is no,” said John Weil, vice president and general manager for IoT and edge AI processor business at Synaptics. “It’s like your laptop today. When you buy a brand new laptop, it gives you 12 hours of battery life, if you’re lucky. If you buy another laptop next year, you get 12 hours of battery life, but it does more. Five years from now, you get 12 hours of battery life, and it does even more again. You’re never going to get 24 hours of battery life. What happens instead is that the laptop or EV features will keep getting better, and they’ll move the bar even further, so the energy and power, and PPA will stay roughly the same. An EV today is much better than an EV three years ago, but it fits in the same chassis.”
For example, in today’s software-defined and autonomous vehicles, AI processing by GPUs or XPUs offers increasing amounts of infotainment and assisted driving features, but this could impact battery charge if the ICs are not sufficiently optimized.
“Power’s a really big thing,” said Rob Fisher, senior director of product management at Imagination Technologies. “Generally, you get to a certain point in power dissipation, and you need to go to a more active cooling approach, maybe a water-cooled system, which costs money. It’s not just the power draw reducing your battery time. It’s also the cost of the system. Performance density and power reduction, i.e., performance per power, is a key metric for these massive amounts of compute.”
Related Articles
Moving Electrons, Not Just Vehicles
Why smarter charging, battery management, and power conversion are now the real differentiators in EVs and edge systems.
AI Power On The Edge
Architecting solutions for edge AI is not about minimizing cloud solutions or making small extensions of existing MCUs/MPUs. It’s a hardware/software/model co-development problem.
New Automotive Architectures Are Shaking Up Processor And Memory Choices
Exponential increases in data and a mix of performance requirements are driving a top-to-bottom rethinking of what works best where.
SDVs And AI Forcing Big Changes In Automotive
Advancements in vehicle architectures, AI, and security are reshaping the future of the car.
 |
 |
 |
 |
 |
 |
 |
 |
Leave a Reply