TI Makes Play for ADAS Radar sensors

By Alan A. Varghese, Contributing Editor.

When a large company enters a new market, the following is usually true:

  1. The market is a high-growth one from the point of view of market trends, end-customer demand, government mandates, and regulations—which can be boiled down to increasing TAMs and SAMs.
  2. The market is undergoing disruption in performance, standards, and technologies, and the company believes it has unique differentiators to offer compared to incumbents.

Automotive radar is clearly a high-growth market, so I was not surprised when Texas Instruments (TI) recently announced its single-chip CMOS offering for 76 to 81 GHz automotive radar applications. Still, I couldn’t help wondering what differentiators TI could offer, considering the heavyweight incumbents in this space: Infineon, NXP, and STMicro.

I spent the ’90s at Ericsson programming digital signal processors (DSPs) like the TI C5X and C54X for its digital cellular handsets. From those DSP roots, TI has morphed into a much different company over the years by adding analog and mixed-signal capabilities. In fact, during TI’s acquisition of analog company Burr-Brown in 2000, TI’s then-CEO Tom Engibous declared that “We are as serious about analog as we are about DSP.”

This expertise is obviously relevant when you consider that the radar signal has to be first received in the analog domain, then converted, and finally digitized in the mixed-signal domain. It then goes through digital signal processing (Fig. 1).

Figure 1 - Optimum radar solutions require end-to-end understanding of analog, mixed-signal, and digital signal processing. (Courtesy of SpringerPlus)

Fig. 1
Optimum radar solutions require end-to-end understanding of analog, mixed-signal, and digital signal processing. (Courtesy of SpringerPlus)

Note that TI’s automotive radar solution will be in CMOS. What are the advantages and disadvantages of such an implementation? CMOS mask-sets in the advanced nodes can cost a few million dollars, which is usually justified only for high-volume markets such as mobile phones (1.5 billion units a year), consumer electronics, broadband modems, etc. Consider, however, that annual volumes in the automobile industry amount to 90-100 million vehicles per year and that each vehicle after year 2021 could have as many as 10 radar units. In doing so, we arrive at volumes that are not too different from those of the mobile phone industry.

The advantage of CMOS is that, once you have amortized the cost of the mask-set, CMOS front-ends can offer tight integration with the baseband and microcontroller sections. They therefore can result in small-form-factor, low-power, low-cost architectures.

In terms of RF, CMOS does have a performance disadvantage. Millimeter-wave circuits are typically designed in high-performance RF technologies due to the limited performance of CMOS transistors. The cutoff frequency and maximum frequency of oscillation (fT, fmax), which are typically between 200 and 300 GHz for deep-scaled CMOS, are not as high as they are for III-V technologies. Thus, achieving good noise figure and gain with low crosstalk between different channels is challenging.

High parasitics, lossy substrate, and the limited accuracy of component models all make the task of CMOS device optimization and modeling difficult. TI claims less than 4-centimeter-range resolution, range accuracy down to less than 50 micrometers, and range up to 300 meters for its CMOS radar solution. But this performance needs to be proven in real-world driving conditions.

The first automotive radar sensors in the 76-to-81-GHz band available in the late 1990s were architected using gallium-arsenide (GaAs) devices or discrete Schottky diodes. The architectures were expensive, and thus radar was relegated primarily to the high-end auto segment. Around 2008, Infineon demonstrated the first silicon-germanium (SiGe) -based 77-GHz transceivers using bare dies that were glued and bonded to the printed circuit board. This was followed in 2012 by the first SiGe monolithic microwave integrated circuit (MMIC), which enabled the penetration of radar into the mid-range auto segments.

Since then, Infineon has implemented automotive radar primarily in SiGe. The latter has superior high-frequency characteristics, such as high fT and fmax, which simplifies design efforts and provides higher-performance margins.

Around 2016, Infineon began working jointly with IMEC in Belgium to develop integrated CMOS-based radar sensor chips for the 79-GHz band. IMEC had been working for a few years prior on 28-nm CMOS radar sensors. In early 2016, NXP also announced initiatives in CMOS. The company’s 7.5 × 7.5-mm CMOS-based radar transceiver chip included the low-noise amplifiers, mixers, power amplifiers, and voltage-controlled oscillators for the 77-GHz band. Due to the miniaturization of the solution, the company claimed that customers could place them in the four corners of the auto, or alternately use a system of 10 to 20 radar chips distributed all around the auto.

For its approach to this market, TI has chosen to combine the RF radar transceiver with the DSP and microcontroller (MCU) all in a single chip (Fig. 2). There are both positives and negatives to this type of architecture. When prior technologies such as GPS, Bluetooth, and Wi-Fi were implemented in single-chip CMOS, they found themselves adopted rapidly into consumer markets. We saw hockey-stick-like market growth curves, which the company clearly hopes to see here. Automotive is just the tip of the spear as far as commercial radar is concerned; other potential commercial markets include building automation, drones, factory automation, robotics, and wearables.

Figure 2 - Texas Instruments single chip CMOS example

Fig. 2
The new strategy may be to leverage TI’s success in single-chip CMOS for automotive radar into multiple future applications in consumer and industrial markets. (Courtesy of Texas Instruments)

The main problem with integrating RF, mixed-signal, and digital domains onto a single chip is that the whole chip will need to be redesigned whenever there are requirement changes in any domain. Also, keep in mind that digital scaling happens on a more rapid pace than RF and analog. By integrating these domains, it burdens digital innovation to be synchronized to RF and analog innovation, which will unfortunately affect both power consumption and size.

MarketsandMarkets projects that automotive radar will grow at a CAGR of about 24%, expanding from $2.27 billion in 2016 to $6.61 billion by 2021. Similarly, analyst firm Hexa Research claims that automotive radar is anticipated to reach nearly $9 billion by 2024. With this kind of market opportunity, it was inevitable that a semiconductor stalwart such as TI would enter the fray. The company has an uphill climb with heavyweight incumbent competitors, but its single-chip CMOS product might be what helps it get its foot in the door.


Start typing and press Enter to search

Ultrasonic sensors in passenger vehiclesThe Xavier SoC integrates a new GPU architecture called Volta, a custom 8 core CPU architecture, and a new computer vision accelerator. It will deliver 20 TOPS (trillion operations per second) of performance, while consuming only 20 watts of power.