The energy industry is in the first phases of the turn of the century. And one of the most important aspects of this shift is that electric vehicles, solar farm power supplies and devices will inherently rely more on digital technologies. As Hamed Heyhat, general manager of grid automation at General Electric, says, “Decarbonization cannot happen without digitizing the grid.”
So how will we see this impact play out in the chip industry?
1. The grid becomes a network
Power grids are considered to be the most complex machines ever built. Utility companies provide real-time, 24/7 power to millions of customers spread over thousands of square miles.
But they are not flexible. Utilities have little insight into your electricity use and even less control over how much you use. For a margin of error, they invest in expensive, underused peaker equipment. California alone has 17 GW of Peaker plans and they are used less than 15% of the time.
The transformation of the power grid into an intelligent, multi-directional network, where the refinement of power loads with AI and processing replaces the brute force of overcapacity, is already underway. Span has developed an intelligent control panel powered by a quad-core Arm processor and AI that effectively turns a home into a microgrid: residents can power their home from their car during natural disasters, or remotely turn off appliances to save money save.
Enphase, meanwhile, has developed its own ARM-based ASICs to power microinverters to boost solar output, control battery systems, and sell excess power to utilities. In New England, utility companies are offering customers up to $1,000 a year for limited access to battery systems on select days. Inverters and optimizers that are “smarter” with CPUs and software have taken anywhere from 30% to over 80% share of the US housing market.
The demand for visibility grows exponentially as you move to distribution and transmission networks. Edge Impulse experiments with computer vision at the edge to detect faults and fires faster, while Awesense builds renewable energy microgrids for corporate campuses. It has set itself the goal of avoiding 100 million tons of CO2 by 2025. Due to the stringent safety requirements in this market, it is expected that some of the design ideas to isolate workloads and/or ensure reliability will increasingly be incorporated into automotive ADAS systems will find their way into these devices.
2nd building too
Critics might argue that equipping every household with a quad-core CPU will not fundamentally increase shipping volume. There are only around 140 million residential homes and nearly 6 million commercial buildings in the US, making the TAM for smart building controls in this country 150 million.
But smart building systems have to talk to something. Appliance manufacturer Arcelik has found that running relatively simple AI algorithms at the edge can reduce refrigerator power consumption by 10%. Used across Europe, this could replace nine power plants. Aquaseca is developing a device that detects pipe leaks using sound detection: Insurers are testing such devices to stem $13 billion in annual water damage claims. The space, power, and thermal limitations of home appliances are also intriguing testbeds for NPUs.
There will also be a fleet of less intelligent, less programmable, but still “aware” devices like lightbulbs (40 per household) and sockets (75). The amounts of MCUs, flash and embedded communications required in buildings will reach billions of units, while the complexity of embedded intelligence will increase over time.
3. Silicon boosts renewables
Most of the costs for solar and wind power systems are incurred up front during the construction of the system. The biggest variable costs are downtime and maintenance.
In wind farms, the main culprits are the bearings that move the nacelle and rotor blades. Running a turbine to failure can cost $150,000 in repairs and lost production as the turbine frequently returns to the shop floor. Onsite preventative repair over IoT and 5G or LTE costs $5,000.
While there are few moving parts in solar fields, preventative repairs via IoT and AI could add $500,000 per year, or $17 million over its lifetime, to the profits of a 50MW solar array. (Added bonus: the CPU in the inverter can be used for many of these tasks). Because of this sensory awareness, utilities like the Arizona Public Service use small teams to manage solar arrays that span thousands of square miles. There’s no shortage of data for AI algorithms to analyze: a 1MW wind farm produces 9 times more data than its fossil counterpart, while a solar array produces 40 times more data.
How big is the chance? The IEA estimates that we will need $4 trillion in clean energy investments annually from 2030 onwards. Solar and wind power are projected to grow from 10% of global capacity to 70% by 2050.
4. The non-CMOS industry charges
The price of battery packs has fallen from around $1,200 per kilowatt hour to $135 over the past decade. A good chunk of the profits came from the magic of mass manufacturing. Bringing the price down to $45 per kWh by 2035 will require a storage transformation.
Silicon Carbine (SiC) and Gallium Nitride (GaN) semiconductors can increase power conversion and efficiency of inverters and SoC. This in turn means more miles for an electric vehicle, more hours of off-grid power in a home, and/or smaller, more affordable battery packs. GaN and SiC can also speed up charging of electric vehicles or reduce transmission and distribution losses in the grid. Deployment is also simplified: a SiC-based solid-state transformer that would fit in a suitcase can replace 8,000 pounds of copper wire in a traditional transformer.
SiC and GaN have been around for years, but the volume requirements are comparatively small. Changing the fuel source changes the picture. Ford and other manufacturers are reportedly signing agreements to lock down capacity.
5. Design and manufacturing become clean
The push toward sustainability also means that the semiconductor industry itself must become more energy efficient. Last year, TSMC signed an agreement to purchase 920MW of offshore wind power to power factories. The company is now one of the ten largest buyers of renewable energies worldwide. (And will disruptions be an issue? Bloomberg New Energy Finance estimates that spare storage capacity in UPSs could provide gigawatts of capacity for the grid.)
In the meantime, we will all find ways to reduce the time, energy, and emissions associated with design, test, and verification. Arm, for example, has started moving many design processes to the cloud. We expect this to help us reduce data center energy consumption by 45% while speeding up tasks by 6x in some cases.
The long and winding road
The energy and utilities industry is a little different than many traditional electronics markets. Technology investments often have to be approved by public commissions. Technology testing can take years. The increased safety awareness also requires suppliers to meet strict requirements. But one thing we do know is that change is coming and Arm, along with the rest of our industry, will play a big part in it.